diff --git a/.github/workflows/agentry-issue-triage.yml b/.github/workflows/agentry-bug-fix.yml
similarity index 61%
rename from .github/workflows/agentry-issue-triage.yml
rename to .github/workflows/agentry-bug-fix.yml
index f3c66d2..4e60b92 100644
--- a/.github/workflows/agentry-issue-triage.yml
+++ b/.github/workflows/agentry-bug-fix.yml
@@ -1,26 +1,29 @@
 # This workflow is a manually refined version of the output generated by:
-#   agentry ci generate --target github --dry-run workflows/triage.yaml
+#   agentry ci generate --target github --dry-run workflows/bug-fix.yaml
 #
 # Manually refined to include:
-# - Proper workflow name capitalization ("Issue Triage" instead of "triage")
-# - Permissions for issues write access (issue:comment and issue:label tools)
+# - Proper workflow name capitalization ("Bug Fix" instead of "bug-fix")
+# - Permissions for issue write and pull-request write access (pr:create and issue:comment tools)
 # - Workflow-level env section with CLAUDE_CODE_OAUTH_TOKEN
 # - Local installation via "pip install ." (using workspace as source)
-# - Specific agentry run parameters: --input issue-description, --input repository-ref, --binder, --output-format
+# - Specific agentry run parameters: --input repository-ref, --binder, --output-format
 # - Improved formatting and readability with proper spacing
+# - Conditional trigger: only runs when the applied label is "category:bug"
 
-name: "Agentry: Issue Triage"
+name: "Agentry: Bug Fix"
 
 'on':
   issues:
-    types: [opened]
+    types: [labeled]
 
 permissions:
-  contents: read
+  contents: write
   issues: write
+  pull-requests: write
 
 jobs:
   agentry:
+    if: github.event.label.name == 'category:bug'
     runs-on: ubuntu-latest
     steps:
       - name: Checkout repository
@@ -39,7 +42,7 @@ jobs:
 
       - name: Run agentry
         run: >
-          agentry --output-format json run workflows/triage.yaml
+          agentry --output-format json run workflows/bug-fix.yaml
           --input repository-ref=.
           --binder github-actions
         env:
diff --git a/.github/workflows/agentry-feature-implement.yml b/.github/workflows/agentry-feature-implement.yml
new file mode 100644
index 0000000..d3fa12d
--- /dev/null
+++ b/.github/workflows/agentry-feature-implement.yml
@@ -0,0 +1,50 @@
+# This workflow is a manually refined version of the output generated by:
+#   agentry ci generate --target github --dry-run workflows/feature-implement.yaml
+#
+# Manually refined to include:
+# - Proper workflow name capitalization ("Feature Implement" instead of "feature-implement")
+# - Permissions for issue write and pull-request write access (pr:create and issue:comment tools)
+# - Workflow-level env section with CLAUDE_CODE_OAUTH_TOKEN
+# - Local installation via "pip install ." (using workspace as source)
+# - Specific agentry run parameters: --input repository-ref, --binder, --output-format
+# - Improved formatting and readability with proper spacing
+# - Conditional trigger: only runs when the applied label is "category:feature"
+
+name: "Agentry: Feature Implement"
+
+'on':
+  issues:
+    types: [labeled]
+
+permissions:
+  contents: write
+  issues: write
+  pull-requests: write
+
+jobs:
+  agentry:
+    if: github.event.label.name == 'category:feature'
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      - name: Install Claude Code
+        run: npm install -g @anthropic-ai/claude-code
+
+      - name: Install agentry
+        run: pip install .
+
+      - name: Run agentry
+        run: >
+          agentry --output-format json run workflows/feature-implement.yaml
+          --input repository-ref=.
+          --binder github-actions
+        env:
+          CLAUDE_CODE_OAUTH_TOKEN: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
diff --git a/.github/workflows/agentry-planning-pipeline.yml b/.github/workflows/agentry-planning-pipeline.yml
new file mode 100644
index 0000000..7bcc1c6
--- /dev/null
+++ b/.github/workflows/agentry-planning-pipeline.yml
@@ -0,0 +1,40 @@
+# This workflow replaces the superseded agentry-issue-triage.yml workflow.
+# It triggers on issue creation and runs the planning pipeline which handles issue triage
+# and task decomposition more comprehensively than the triage-only workflow.
+
+name: "Agentry: Planning Pipeline"
+
+'on':
+  issues:
+    types: [opened]
+
+permissions:
+  contents: read
+  issues: write
+
+jobs:
+  agentry:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      - name: Install Claude Code
+        run: npm install -g @anthropic-ai/claude-code
+
+      - name: Install agentry
+        run: pip install .
+
+      - name: Run agentry
+        run: >
+          agentry --output-format json run workflows/planning-pipeline.yaml
+          --input repository-ref=.
+          --binder github-actions
+        env:
+          CLAUDE_CODE_OAUTH_TOKEN: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
diff --git a/docs/specs/04-spec-agentry-ci/02-proofs/T02-01-test.txt b/docs/specs/04-spec-agentry-ci/02-proofs/T02-01-test.txt
deleted file mode 100644
index b7d42dc..0000000
--- a/docs/specs/04-spec-agentry-ci/02-proofs/T02-01-test.txt
+++ /dev/null
@@ -1,22 +0,0 @@
-Type: test
-Command: uv run pytest tests/unit/test_github_binder_inputs.py::TestResolveInputsIssueBodySource -v
-Expected: All 9 new T02 tests pass
-Timestamp: 2026-03-27T00:00:00Z
-
-============================= test session starts ==============================
-platform linux -- Python 3.13.7, pytest-9.0.2, pluggy-1.6.0
-configfile: pyproject.toml
-
-tests/unit/test_github_binder_inputs.py::TestResolveInputsIssueBodySource::test_resolves_issue_body_from_issues_event PASSED
-tests/unit/test_github_binder_inputs.py::TestResolveInputsIssueBodySource::test_falls_back_to_issue_title_when_body_is_null PASSED
-tests/unit/test_github_binder_inputs.py::TestResolveInputsIssueBodySource::test_falls_back_to_issue_title_when_body_is_absent PASSED
-tests/unit/test_github_binder_inputs.py::TestResolveInputsIssueBodySource::test_falls_back_to_issue_title_when_body_is_empty_string PASSED
-tests/unit/test_github_binder_inputs.py::TestResolveInputsIssueBodySource::test_fallback_emits_warning_log PASSED
-tests/unit/test_github_binder_inputs.py::TestResolveInputsIssueBodySource::test_cli_provided_value_overrides_source_and_fallback PASSED
-tests/unit/test_github_binder_inputs.py::TestResolveInputsIssueBodySource::test_cli_override_takes_precedence_even_when_body_null PASSED
-tests/unit/test_github_binder_inputs.py::TestResolveInputsIssueBodySource::test_no_fallback_returns_none_when_source_missing PASSED
-tests/unit/test_github_binder_inputs.py::TestResolveInputsIssueBodySource::test_fallback_not_triggered_when_source_resolves PASSED
-
-============================== 9 passed in 0.03s ===============================
-
-Status: PASS
diff --git a/docs/specs/04-spec-agentry-ci/02-proofs/T02-02-test.txt b/docs/specs/04-spec-agentry-ci/02-proofs/T02-02-test.txt
deleted file mode 100644
index 83f71dd..0000000
--- a/docs/specs/04-spec-agentry-ci/02-proofs/T02-02-test.txt
+++ /dev/null
@@ -1,10 +0,0 @@
-Type: test
-Command: uv run pytest tests/unit/ -q
-Expected: Full unit test suite passes with no regressions
-Timestamp: 2026-03-27T00:00:00Z
-
-1572 passed, 1 skipped in 8.20s
-
-(T02 adds 9 new tests to test_github_binder_inputs.py; total includes tests from concurrent T01 worker)
-
-Status: PASS
diff --git a/docs/specs/04-spec-agentry-ci/02-proofs/T02-proofs.md b/docs/specs/04-spec-agentry-ci/02-proofs/T02-proofs.md
deleted file mode 100644
index 2f84dd7..0000000
--- a/docs/specs/04-spec-agentry-ci/02-proofs/T02-proofs.md
+++ /dev/null
@@ -1,45 +0,0 @@
-# T02 Proof Summary: Issue Input Resolution via Source Mapping
-
-## Task
-T02 — Enable the triage workflow's `issue-description` string input to resolve from
-the GitHub issue event payload using existing source mapping.
-
-## Changes Implemented
-
-### 1. `workflows/triage.yaml`
-- Added `source: issue.body` field to the `issue-description` input declaration
-- Added `fallback: issue.title` field so that when `issue.body` is null/empty, the
-  issue title is used as a fallback with a warning log
-
-### 2. `src/agentry/binders/github_actions.py`
-- Added `import logging` and `logger = logging.getLogger(__name__)`
-- Updated `_resolve_string()` to support the `fallback` key in input specs:
-  - When `source` resolves to null or empty string, the `fallback` dotpath is tried
-  - A `WARNING` level log is emitted when falling back, including the input name,
-    source path, fallback path, and resolved fallback value
-  - CLI `--input` overrides (via `provided_values`) still take strict priority
-    over both `source` and `fallback`
-
-### 3. `tests/unit/test_github_binder_inputs.py`
-- Added `TestResolveInputsIssueBodySource` class with 9 new tests covering:
-  - `issue.body` resolution from `issues` event payload
-  - Fallback to `issue.title` when body is null
-  - Fallback to `issue.title` when body key is absent
-  - Fallback to `issue.title` when body is an empty string
-  - Warning log emission on fallback
-  - CLI `--input` override taking precedence over source and fallback
-  - CLI override wins even when body is null
-  - No fallback key: returns None for optional missing source
-  - Fallback not triggered when source resolves to a non-empty value
-
-## Proof Artifacts
-
-| File | Type | Status |
-|------|------|--------|
-| T02-01-test.txt | test (new tests only) | PASS |
-| T02-02-test.txt | test (full unit suite) | PASS |
-
-## Test Counts
-- Baseline: 1534 passed, 1 skipped
-- After T02: 1572 passed, 1 skipped (includes T01's tests too)
-- T02 adds 9 new tests in `TestResolveInputsIssueBodySource`
diff --git a/docs/specs/07-spec-issue-triage/01-proofs/T01-01-test.txt b/docs/specs/07-spec-issue-triage/01-proofs/T01-01-test.txt
deleted file mode 100644
index 1b5a611..0000000
--- a/docs/specs/07-spec-issue-triage/01-proofs/T01-01-test.txt
+++ /dev/null
@@ -1,113 +0,0 @@
-Type: test
-Command: uv run pytest tests/unit/test_github_binder_tools.py tests/integration/test_issue_tools.py -v
-Expected: All 97 tests pass
-Timestamp: 2026-03-27T00:00:00Z
-
-============================= test session starts ==============================
-platform linux -- Python 3.13.7, pytest-9.0.2, pluggy-1.6.0 -- <workspace>/.venv/bin/python
-cachedir: .pytest_cache
-rootdir: <workspace>
-configfile: pyproject.toml
-plugins: asyncio-1.3.0, cov-7.0.0, anyio-4.12.1
-asyncio: mode=Mode.STRICT, debug=False, asyncio_default_fixture_loop_scope=None, asyncio_default_test_loop_scope=function
-collecting ... collected 97 items
-
-tests/unit/test_github_binder_tools.py::TestBindToolsSupported::test_all_supported_tools_can_be_bound PASSED [  1%]
-tests/unit/test_github_binder_tools.py::TestBindToolsSupported::test_bound_tools_are_callable PASSED [  2%]
-tests/unit/test_github_binder_tools.py::TestBindToolsSupported::test_empty_tool_list_returns_empty_dict PASSED [  3%]
-tests/unit/test_github_binder_tools.py::TestBindToolsSupported::test_bind_single_tool_repository_read PASSED [  4%]
-tests/unit/test_github_binder_tools.py::TestBindToolsSupported::test_bind_single_tool_shell_execute PASSED [  5%]
-tests/unit/test_github_binder_tools.py::TestBindToolsRepositoryRead::test_reads_file_from_workspace PASSED [  6%]
-tests/unit/test_github_binder_tools.py::TestBindToolsRepositoryRead::test_reads_nested_file_from_workspace PASSED [  7%]
-tests/unit/test_github_binder_tools.py::TestBindToolsRepositoryRead::test_path_traversal_raises_error PASSED [  8%]
-tests/unit/test_github_binder_tools.py::TestBindToolsRepositoryRead::test_path_traversal_error_message_contains_path PASSED [  9%]
-tests/unit/test_github_binder_tools.py::TestBindToolsRepositoryRead::test_workspace_root_used_not_cwd PASSED [ 10%]
-tests/unit/test_github_binder_tools.py::TestBindToolsRepositoryRead::test_missing_file_raises_file_not_found PASSED [ 11%]
-tests/unit/test_github_binder_tools.py::TestBindToolsRepositoryRead::test_requires_path_kwarg PASSED [ 12%]
-tests/unit/test_github_binder_tools.py::TestBindToolsShellExecute::test_allowed_command_git_log_executes PASSED [ 13%]
-tests/unit/test_github_binder_tools.py::TestBindToolsShellExecute::test_disallowed_command_rm_raises_error PASSED [ 14%]
-tests/unit/test_github_binder_tools.py::TestBindToolsShellExecute::test_disallowed_command_curl_raises_error PASSED [ 15%]
-tests/unit/test_github_binder_tools.py::TestBindToolsShellExecute::test_disallowed_command_error_message_contains_command PASSED [ 16%]
-tests/unit/test_github_binder_tools.py::TestBindToolsShellExecute::test_disallowed_git_subcommand_raises_error PASSED [ 17%]
-tests/unit/test_github_binder_tools.py::TestBindToolsShellExecute::test_allowed_ls_command PASSED [ 18%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRComment::test_pr_comment_posts_to_correct_url PASSED [ 19%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRComment::test_pr_comment_uses_github_api_base_url PASSED [ 20%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRComment::test_pr_comment_posts_body_in_payload PASSED [ 21%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRComment::test_pr_comment_uses_auth_token PASSED [ 22%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRComment::test_pr_comment_returns_api_response PASSED [ 23%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRComment::test_pr_comment_on_non_pr_event_raises_value_error PASSED [ 24%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRComment::test_pr_comment_requires_body_kwarg PASSED [ 25%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRReview::test_pr_review_posts_to_correct_url PASSED [ 26%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRReview::test_pr_review_uses_github_api_base_url PASSED [ 27%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRReview::test_pr_review_posts_body_in_payload PASSED [ 28%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRReview::test_pr_review_includes_event_in_payload PASSED [ 29%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRReview::test_pr_review_default_event_is_comment PASSED [ 30%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRReview::test_pr_review_uses_auth_token PASSED [ 31%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRReview::test_pr_review_on_non_pr_event_raises_value_error PASSED [ 32%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRReview::test_pr_review_returns_api_response PASSED [ 34%]
-tests/unit/test_github_binder_tools.py::TestBindToolsPRReview::test_pr_review_accepts_inline_comments PASSED [ 35%]
-tests/unit/test_github_binder_tools.py::TestBindToolsUnsupportedTools::test_unknown_tool_raises_unsupported_tool_error PASSED [ 36%]
-tests/unit/test_github_binder_tools.py::TestBindToolsUnsupportedTools::test_error_message_contains_tool_name PASSED [ 37%]
-tests/unit/test_github_binder_tools.py::TestBindToolsUnsupportedTools::test_unknown_tool_in_mixed_list_raises PASSED [ 38%]
-tests/unit/test_github_binder_tools.py::TestBindToolsUnsupportedTools::test_error_mentions_unsupported_tool_name PASSED [ 39%]
-tests/unit/test_github_binder_tools.py::TestBindToolsUnsupportedTools::test_arbitrary_unsupported_tool_name PASSED [ 40%]
-tests/unit/test_github_binder_tools.py::TestPRCommentAPIErrors::test_403_error_mentions_status_code PASSED [ 41%]
-tests/unit/test_github_binder_tools.py::TestPRCommentAPIErrors::test_403_error_includes_scope_remediation PASSED [ 42%]
-tests/unit/test_github_binder_tools.py::TestPRCommentAPIErrors::test_404_error_mentions_status_code PASSED [ 43%]
-tests/unit/test_github_binder_tools.py::TestPRCommentAPIErrors::test_404_error_includes_pr_number PASSED [ 44%]
-tests/unit/test_github_binder_tools.py::TestPRCommentAPIErrors::test_network_timeout_raises_runtime_error PASSED [ 45%]
-tests/unit/test_github_binder_tools.py::TestPRCommentAPIErrors::test_500_error_raises_runtime_error_with_status PASSED [ 46%]
-tests/unit/test_github_binder_tools.py::TestPRReviewAPIErrors::test_403_error_includes_scope_remediation PASSED [ 47%]
-tests/unit/test_github_binder_tools.py::TestPRReviewAPIErrors::test_404_error_includes_pr_not_found_info PASSED [ 48%]
-tests/unit/test_github_binder_tools.py::TestPRReviewAPIErrors::test_network_timeout_raises_runtime_error PASSED [ 49%]
-tests/unit/test_github_binder_tools.py::TestExtractIssueNumber::test_returns_none_for_non_issues_event PASSED [ 50%]
-tests/unit/test_github_binder_tools.py::TestExtractIssueNumber::test_returns_none_for_pull_request_event PASSED [ 51%]
-tests/unit/test_github_binder_tools.py::TestExtractIssueNumber::test_returns_issue_number_for_issues_event PASSED [ 52%]
-tests/unit/test_github_binder_tools.py::TestExtractIssueNumber::test_returns_none_when_issue_key_absent PASSED [ 53%]
-tests/unit/test_github_binder_tools.py::TestExtractIssueNumber::test_returns_none_when_number_key_absent PASSED [ 54%]
-tests/unit/test_github_binder_tools.py::TestExtractIssueNumber::test_converts_number_to_int PASSED [ 55%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueComment::test_issue_comment_posts_to_correct_url PASSED [ 56%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueComment::test_issue_comment_uses_github_api_base_url PASSED [ 57%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueComment::test_issue_comment_posts_body_in_payload PASSED [ 58%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueComment::test_issue_comment_uses_auth_token PASSED [ 59%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueComment::test_issue_comment_returns_api_response PASSED [ 60%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueComment::test_issue_comment_on_non_issues_event_raises_value_error PASSED [ 61%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueComment::test_issue_comment_on_pr_event_raises_value_error PASSED [ 62%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueComment::test_issue_comment_requires_body_kwarg PASSED [ 63%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueComment::test_issue_comment_403_error_mentions_issues_scope PASSED [ 64%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueComment::test_issue_comment_404_error_mentions_issue_number PASSED [ 65%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueComment::test_issue_comment_timeout_raises_runtime_error PASSED [ 67%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueLabel::test_issue_label_posts_to_correct_url PASSED [ 68%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueLabel::test_issue_label_uses_github_api_base_url PASSED [ 69%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueLabel::test_issue_label_posts_labels_in_payload PASSED [ 70%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueLabel::test_issue_label_uses_auth_token PASSED [ 71%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueLabel::test_issue_label_returns_api_response PASSED [ 72%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueLabel::test_issue_label_on_non_issues_event_raises_value_error PASSED [ 73%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueLabel::test_issue_label_on_pr_event_raises_value_error PASSED [ 74%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueLabel::test_issue_label_requires_labels_kwarg PASSED [ 75%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueLabel::test_issue_label_403_error_mentions_issues_scope PASSED [ 76%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueLabel::test_issue_label_404_error_mentions_issue_number PASSED [ 77%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueLabel::test_issue_label_422_error_mentions_validation PASSED [ 78%]
-tests/unit/test_github_binder_tools.py::TestBindToolsIssueLabel::test_issue_label_timeout_raises_runtime_error PASSED [ 79%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_sends_correct_api_request PASSED [ 80%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_sends_correct_body PASSED [ 81%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_includes_authorization_header PASSED [ 82%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_returns_response_json PASSED [ 83%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_403_raises_runtime_with_remediation PASSED [ 84%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_404_raises_runtime_with_issue_number PASSED [ 85%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_timeout_raises_runtime_error PASSED [ 86%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_raises_on_non_issues_event PASSED [ 87%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_sends_correct_api_request PASSED [ 88%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_sends_labels_in_payload PASSED [ 89%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_includes_authorization_header PASSED [ 90%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_returns_response_json PASSED [ 91%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_403_raises_runtime_with_remediation PASSED [ 92%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_404_raises_runtime_with_issue_number PASSED [ 93%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_422_raises_runtime_with_validation_hint PASSED [ 94%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_timeout_raises_runtime_error PASSED [ 95%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_raises_on_non_issues_event PASSED [ 96%]
-tests/integration/test_issue_tools.py::TestSupportedToolsContainsIssueTools::test_issue_comment_in_supported_tools PASSED [ 97%]
-tests/integration/test_issue_tools.py::TestSupportedToolsContainsIssueTools::test_issue_label_in_supported_tools PASSED [ 98%]
-tests/integration/test_issue_tools.py::TestSupportedToolsContainsIssueTools::test_both_issue_tools_can_be_bound_together PASSED [100%]
-
-============================== 97 passed in 0.09s ==============================
diff --git a/docs/specs/07-spec-issue-triage/01-proofs/T01-02-cli.txt b/docs/specs/07-spec-issue-triage/01-proofs/T01-02-cli.txt
deleted file mode 100644
index 526cd37..0000000
--- a/docs/specs/07-spec-issue-triage/01-proofs/T01-02-cli.txt
+++ /dev/null
@@ -1,7 +0,0 @@
-Type: cli
-Command: python -c "from agentry.binders.github_actions import SUPPORTED_TOOLS; print(sorted(SUPPORTED_TOOLS))"
-Expected: 'issue:comment' and 'issue:label' present in SUPPORTED_TOOLS
-Timestamp: 2026-03-27T00:00:00Z
-Status: PASS
-
-['issue:comment', 'issue:label', 'pr:comment', 'pr:create', 'pr:review', 'repository:read', 'shell:execute']
diff --git a/docs/specs/07-spec-issue-triage/01-proofs/T01-proofs.md b/docs/specs/07-spec-issue-triage/01-proofs/T01-proofs.md
deleted file mode 100644
index ac5feed..0000000
--- a/docs/specs/07-spec-issue-triage/01-proofs/T01-proofs.md
+++ /dev/null
@@ -1,54 +0,0 @@
-# T01 Proof Summary: Issue Tool Bindings in GitHubActionsBinder
-
-## Task
-
-Add `issue:comment` and `issue:label` tool bindings to `GitHubActionsBinder` so
-workflows can interact with GitHub issues.
-
-## Changes Made
-
-**Modified:**
-- `src/agentry/binders/github_actions.py`
-  - Extended `SUPPORTED_TOOLS` frozenset with `"issue:comment"` and `"issue:label"`
-  - Added `_extract_issue_number()` static method (parallel to `_extract_pr_number`)
-  - Store `self._issue_number` in `__init__` from issues event payload
-  - Implemented `_make_issue_comment()` — POST `/repos/{owner}/{repo}/issues/{number}/comments`
-  - Implemented `_make_issue_label()` — POST `/repos/{owner}/{repo}/issues/{number}/labels`
-  - Wired both in `bind_tools()` dispatch
-
-**Modified:**
-- `tests/unit/test_github_binder_tools.py`
-  - Added `TestExtractIssueNumber` class (6 tests)
-  - Added `TestBindToolsIssueComment` class (11 tests)
-  - Added `TestBindToolsIssueLabel` class (12 tests)
-
-**Created:**
-- `tests/integration/test_issue_tools.py`
-  - `TestIssueCommentIntegration` (8 tests)
-  - `TestIssueLabelIntegration` (9 tests)
-  - `TestSupportedToolsContainsIssueTools` (3 tests)
-
-## Proof Artifacts
-
-| Artifact | Type | Status |
-|----------|------|--------|
-| T01-01-test.txt | test | PASS (97/97) |
-| T01-02-cli.txt | cli | PASS |
-
-## Test Results
-
-- 97 tests total, 97 passed, 0 failed
-- All existing binder tests continue to pass (48 pre-existing)
-- 49 new tests covering issue:comment and issue:label bindings
-
-## Key Behaviours Verified
-
-1. `issue:comment` and `issue:label` appear in `SUPPORTED_TOOLS`
-2. `_extract_issue_number()` returns the issue number from `issues` events, `None` otherwise
-3. `_issue_number` is `None` when event is not `issues`
-4. `issue:comment` posts to `POST /repos/{owner}/{repo}/issues/{number}/comments`
-5. `issue:label` posts to `POST /repos/{owner}/{repo}/issues/{number}/labels`
-6. Both tools raise `ValueError` with "issues event" message when called outside an issues context
-7. Both tools raise `RuntimeError` with structured remediation on 403/404 API errors
-8. Both tools raise `RuntimeError` with "timeout" on network timeout
-9. `issue:label` additionally handles 422 (validation failure) with remediation hint
diff --git a/docs/specs/07-spec-issue-triage/03-proofs/T03-01-test.txt b/docs/specs/07-spec-issue-triage/03-proofs/T03-01-test.txt
deleted file mode 100644
index 0ef5f83..0000000
--- a/docs/specs/07-spec-issue-triage/03-proofs/T03-01-test.txt
+++ /dev/null
@@ -1,46 +0,0 @@
-============================= test session starts ==============================
-platform linux -- Python 3.13.7, pytest-9.0.2, pluggy-1.6.0 -- <workspace>/.venv/bin/python
-cachedir: .pytest_cache
-rootdir: <workspace>
-configfile: pyproject.toml
-plugins: asyncio-1.3.0, cov-7.0.0, anyio-4.12.1
-asyncio: mode=Mode.STRICT, debug=False, asyncio_default_fixture_loop_scope=None, asyncio_default_test_loop_scope=function
-collecting ... collected 35 items
-
-tests/unit/test_issue_output_formatting.py::TestFormatTriageComment::test_includes_severity_badge_for_critical PASSED [  2%]
-tests/unit/test_issue_output_formatting.py::TestFormatTriageComment::test_includes_severity_badge_for_high PASSED [  5%]
-tests/unit/test_issue_output_formatting.py::TestFormatTriageComment::test_includes_severity_badge_for_medium PASSED [  8%]
-tests/unit/test_issue_output_formatting.py::TestFormatTriageComment::test_includes_severity_badge_for_low PASSED [ 11%]
-tests/unit/test_issue_output_formatting.py::TestFormatTriageComment::test_includes_category PASSED [ 14%]
-tests/unit/test_issue_output_formatting.py::TestFormatTriageComment::test_includes_affected_components PASSED [ 17%]
-tests/unit/test_issue_output_formatting.py::TestFormatTriageComment::test_includes_recommended_assignee PASSED [ 20%]
-tests/unit/test_issue_output_formatting.py::TestFormatTriageComment::test_includes_reasoning PASSED [ 22%]
-tests/unit/test_issue_output_formatting.py::TestFormatTriageComment::test_includes_token_usage_when_present PASSED [ 25%]
-tests/unit/test_issue_output_formatting.py::TestFormatTriageComment::test_fallback_when_output_json_missing PASSED [ 28%]
-tests/unit/test_issue_output_formatting.py::TestFormatTriageComment::test_fallback_for_invalid_json PASSED [ 31%]
-tests/unit/test_issue_output_formatting.py::TestFormatTriageComment::test_fallback_to_raw_response_when_no_structured_data PASSED [ 34%]
-tests/unit/test_issue_output_formatting.py::TestFormatTriageComment::test_full_triage_output_renders_all_sections PASSED [ 37%]
-tests/unit/test_issue_output_formatting.py::TestPostIssueComment::test_posts_to_correct_endpoint PASSED [ 40%]
-tests/unit/test_issue_output_formatting.py::TestPostIssueComment::test_includes_authorization_header PASSED [ 42%]
-tests/unit/test_issue_output_formatting.py::TestPostIssueComment::test_includes_body_in_payload PASSED [ 45%]
-tests/unit/test_issue_output_formatting.py::TestPostIssueComment::test_returns_response_json PASSED [ 48%]
-tests/unit/test_issue_output_formatting.py::TestPostIssueComment::test_403_raises_runtime_with_issues_write_remediation PASSED [ 51%]
-tests/unit/test_issue_output_formatting.py::TestPostIssueComment::test_404_raises_runtime_with_issue_number PASSED [ 54%]
-tests/unit/test_issue_output_formatting.py::TestPostIssueComment::test_timeout_raises_runtime_error PASSED [ 57%]
-tests/unit/test_issue_output_formatting.py::TestPostIssueComment::test_other_http_error_includes_status PASSED [ 60%]
-tests/unit/test_issue_output_formatting.py::TestApplyTriageLabels::test_applies_severity_and_category_labels PASSED [ 62%]
-tests/unit/test_issue_output_formatting.py::TestApplyTriageLabels::test_applies_only_severity_when_category_missing PASSED [ 65%]
-tests/unit/test_issue_output_formatting.py::TestApplyTriageLabels::test_posts_to_correct_labels_endpoint PASSED [ 68%]
-tests/unit/test_issue_output_formatting.py::TestApplyTriageLabels::test_label_api_error_does_not_propagate PASSED [ 71%]
-tests/unit/test_issue_output_formatting.py::TestApplyTriageLabels::test_label_timeout_does_not_propagate PASSED [ 74%]
-tests/unit/test_issue_output_formatting.py::TestApplyTriageLabels::test_logs_warning_when_output_missing PASSED [ 77%]
-tests/unit/test_issue_output_formatting.py::TestApplyTriageLabels::test_logs_warning_when_output_malformed PASSED [ 80%]
-tests/unit/test_issue_output_formatting.py::TestApplyTriageLabels::test_logs_warning_when_no_severity_or_category PASSED [ 82%]
-tests/unit/test_issue_output_formatting.py::TestMapOutputsIssuesEvent::test_issues_event_posts_triage_comment PASSED [ 85%]
-tests/unit/test_issue_output_formatting.py::TestMapOutputsIssuesEvent::test_issues_event_posts_to_correct_comment_url PASSED [ 88%]
-tests/unit/test_issue_output_formatting.py::TestMapOutputsIssuesEvent::test_issues_event_attempts_label_application PASSED [ 91%]
-tests/unit/test_issue_output_formatting.py::TestMapOutputsIssuesEvent::test_issues_event_label_failure_does_not_fail_map_outputs PASSED [ 94%]
-tests/unit/test_issue_output_formatting.py::TestMapOutputsIssuesEvent::test_issues_event_returns_correct_paths PASSED [ 97%]
-tests/unit/test_issue_output_formatting.py::TestMapOutputsIssuesEvent::test_non_issues_event_does_not_post_triage_comment PASSED [100%]
-
-============================== 35 passed in 0.04s ==============================
diff --git a/docs/specs/07-spec-issue-triage/03-proofs/T03-02-test.txt b/docs/specs/07-spec-issue-triage/03-proofs/T03-02-test.txt
deleted file mode 100644
index ecdc7b5..0000000
--- a/docs/specs/07-spec-issue-triage/03-proofs/T03-02-test.txt
+++ /dev/null
@@ -1,20 +0,0 @@
-============================= test session starts ==============================
-platform linux -- Python 3.13.7, pytest-9.0.2, pluggy-1.6.0 -- <workspace>/.venv/bin/python
-cachedir: .pytest_cache
-rootdir: <workspace>
-configfile: pyproject.toml
-plugins: asyncio-1.3.0, cov-7.0.0, anyio-4.12.1
-asyncio: mode=Mode.STRICT, debug=False, asyncio_default_fixture_loop_scope=None, asyncio_default_test_loop_scope=function
-collecting ... collected 9 items
-
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_complete_triage_output_posts_comment_and_labels PASSED [ 11%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_comment_body_contains_all_triage_fields PASSED [ 22%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_labels_posted_with_correct_format PASSED [ 33%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_label_api_failure_does_not_abort_run PASSED [ 44%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_label_timeout_does_not_abort_run PASSED [ 55%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_comment_uses_correct_endpoint_for_issue PASSED [ 66%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_output_missing_posts_fallback_comment PASSED [ 77%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_runs_dir_created_for_issues_event PASSED [ 88%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_token_usage_included_in_comment PASSED [100%]
-
-============================== 9 passed in 0.03s ===============================
diff --git a/docs/specs/07-spec-issue-triage/03-proofs/T03-03-lint.txt b/docs/specs/07-spec-issue-triage/03-proofs/T03-03-lint.txt
deleted file mode 100644
index 8539cd2..0000000
--- a/docs/specs/07-spec-issue-triage/03-proofs/T03-03-lint.txt
+++ /dev/null
@@ -1 +0,0 @@
-Success: no issues found in 1 source file
diff --git a/docs/specs/07-spec-issue-triage/03-proofs/T03-proofs.md b/docs/specs/07-spec-issue-triage/03-proofs/T03-proofs.md
deleted file mode 100644
index dbb3d7c..0000000
--- a/docs/specs/07-spec-issue-triage/03-proofs/T03-proofs.md
+++ /dev/null
@@ -1,54 +0,0 @@
-# T03 Proof Summary: Triage Output Formatting and Label Derivation
-
-**Task:** T03 - Triage Output Formatting and Label Derivation
-**Timestamp:** 2026-03-27
-**Status:** PASS
-
-## Implementation Summary
-
-Extended `GitHubActionsBinder` in `src/agentry/binders/github_actions.py` with:
-
-1. **`map_outputs()` updated** to detect issues events (`self._issue_number is not None`)
-   and call `_format_triage_comment()`, `_post_issue_comment()`, and `_apply_triage_labels()`.
-
-2. **`_format_triage_comment(output_path)`** - Renders triage agent output as Markdown with:
-   - Severity badge (shield.io badge for critical/high/medium/low)
-   - Category field
-   - Affected components list
-   - Recommended assignee
-   - Reasoning section
-   - Token usage footer
-   - Graceful fallbacks for missing/malformed output.json
-
-3. **`_post_issue_comment(body)`** - Posts the formatted comment to the GitHub Issues API
-   (`POST /repos/{owner}/{repo}/issues/{number}/comments`) with structured error handling
-   for 403, 404, and network timeouts.
-
-4. **`_apply_triage_labels(output_path)`** - Reads severity and category from agent output,
-   applies labels as `severity:{value}` and `category:{value}` to the issue via the GitHub API.
-   Label application is best-effort: all errors are logged as warnings and not propagated.
-
-## Proof Artifacts
-
-| File | Type | Status |
-|------|------|--------|
-| T03-01-test.txt | Unit tests (35 tests) | PASS |
-| T03-02-test.txt | Integration tests (9 tests) | PASS |
-| T03-03-lint.txt | Ruff + mypy on github_actions.py | PASS |
-
-## Files Modified
-
-- `src/agentry/binders/github_actions.py` - Added `_format_triage_comment()`, `_post_issue_comment()`, `_apply_triage_labels()` methods; updated `map_outputs()` to handle issues events.
-
-## Files Created
-
-- `tests/unit/test_issue_output_formatting.py` - 35 unit tests covering all new methods
-- `tests/integration/test_issue_triage_pipeline.py` - 9 integration tests covering end-to-end pipeline flow
-
-## Test Results
-
-- Unit tests: 35 passed
-- Integration tests: 9 passed
-- Full suite: 1694 passed, 3 skipped
-- Ruff: All checks passed
-- mypy: No issues found
diff --git a/docs/specs/07-spec-issue-triage/05-proofs/T05-01-test.txt b/docs/specs/07-spec-issue-triage/05-proofs/T05-01-test.txt
deleted file mode 100644
index 36a4edc..0000000
--- a/docs/specs/07-spec-issue-triage/05-proofs/T05-01-test.txt
+++ /dev/null
@@ -1,19 +0,0 @@
-Type: test
-Command: uv run pytest tests/e2e/test_public_surface.py::test_validate_all_standard_workflows -v
-Expected: All 5 standard workflows validate (including triage.yaml)
-Timestamp: 2026-03-27T00:00:00Z
-
-============================= test session starts ==============================
-platform linux -- Python 3.13.7, pytest-9.0.2, pluggy-1.6.0
-rootdir: <workspace>
-collected 5 items
-
-tests/e2e/test_public_surface.py::test_validate_all_standard_workflows[code-review.yaml] PASSED [ 20%]
-tests/e2e/test_public_surface.py::test_validate_all_standard_workflows[triage.yaml] PASSED [ 40%]
-tests/e2e/test_public_surface.py::test_validate_all_standard_workflows[bug-fix.yaml] PASSED [ 60%]
-tests/e2e/test_public_surface.py::test_validate_all_standard_workflows[task-decompose.yaml] PASSED [ 80%]
-tests/e2e/test_public_surface.py::test_validate_all_standard_workflows[planning-pipeline.yaml] PASSED [100%]
-
-============================== 5 passed in 0.39s ===============================
-
-Status: PASS
diff --git a/docs/specs/07-spec-issue-triage/05-proofs/T05-02-test.txt b/docs/specs/07-spec-issue-triage/05-proofs/T05-02-test.txt
deleted file mode 100644
index 92a2640..0000000
--- a/docs/specs/07-spec-issue-triage/05-proofs/T05-02-test.txt
+++ /dev/null
@@ -1,14 +0,0 @@
-Type: test
-Command: uv run pytest tests/e2e/test_public_surface.py -v -q
-Expected: All 38 e2e tests pass (previously 3 were failing)
-Timestamp: 2026-03-27T00:00:00Z
-
-38 e2e tests collected and run.
-All 38 passed in 45.02s.
-
-Previously failing tests that now PASS:
-  - test_validate_all_standard_workflows[triage.yaml]  (StringInput source/fallback now accepted)
-  - test_run_planning_pipeline_stub  (triage.yaml now loads; max_iterations propagated correctly)
-  - test_run_node_isolation  (triage.yaml now loads for composition sub-workflow)
-
-Status: PASS
diff --git a/docs/specs/07-spec-issue-triage/05-proofs/T05-03-lint.txt b/docs/specs/07-spec-issue-triage/05-proofs/T05-03-lint.txt
deleted file mode 100644
index facf7bb..0000000
--- a/docs/specs/07-spec-issue-triage/05-proofs/T05-03-lint.txt
+++ /dev/null
@@ -1,9 +0,0 @@
-Type: cli
-Command: uv run ruff check src/agentry/ && uv run mypy src/agentry/ --ignore-missing-imports
-Expected: No lint or type errors
-Timestamp: 2026-03-27T00:00:00Z
-
-ruff check src/agentry/: All checks passed!
-mypy src/agentry/ --ignore-missing-imports: Success: no issues found in 58 source files
-
-Status: PASS
diff --git a/docs/specs/07-spec-issue-triage/05-proofs/T05-proofs.md b/docs/specs/07-spec-issue-triage/05-proofs/T05-proofs.md
deleted file mode 100644
index 7c5a822..0000000
--- a/docs/specs/07-spec-issue-triage/05-proofs/T05-proofs.md
+++ /dev/null
@@ -1,42 +0,0 @@
-# T05 Proof Summary: Add source and fallback fields to StringInput model
-
-## Task
-
-Fix validation failure: `StringInput` in `src/agentry/models/inputs.py` uses
-`extra="forbid"` and rejected the `source` and `fallback` fields added to
-`triage.yaml` by T02.
-
-## Changes Made
-
-### Primary Fix (in scope)
-- `src/agentry/models/inputs.py`: Added `source: str | None = None` and
-  `fallback: str | None = None` to `StringInput` model.
-
-### Enabling Changes (needed for tests to pass)
-- `src/agentry/binders/local.py`: Added stub implementations for `issue:comment`
-  and `issue:label` in the local binder so that triage.yaml (which declares
-  those tools) can be executed locally.
-- `workflows/triage.yaml`: Converted from `model:` block to `agent:` block with
-  `max_iterations: 1` so the workflow completes within the 30-second test timeout.
-- `workflows/task-decompose.yaml`: Same conversion for the sub-workflow used by
-  planning-pipeline.
-- `src/agentry/composition/engine.py`: Fixed `max_iterations` propagation from
-  `AgentBlock` to `agent_cfg` when executing composition nodes.
-- `src/agentry/runners/in_process.py`: Fixed `max_iterations` propagation from
-  `agent_config.agent_config` to `AgentTask`.
-- `src/agentry/agents/claude_code.py`: Fixed `_build_command` to use
-  `agent_task.max_iterations` (task-level) over `self._max_turns` (instance-level).
-
-## Proof Artifacts
-
-| File | Type | Status |
-|------|------|--------|
-| T05-01-test.txt | test (validate all standard workflows) | PASS |
-| T05-02-test.txt | test (full e2e suite) | PASS |
-| T05-03-lint.txt | cli (ruff + mypy) | PASS |
-
-## Test Results Summary
-
-- Before: 3 e2e failures (validate[triage.yaml], planning_pipeline_stub, node_isolation)
-- After: 0 e2e failures (38/38 pass)
-- Unit/integration: 1694 passed, 3 skipped
diff --git a/docs/specs/triage/04-proofs/T04-01-workflow-validation.txt b/docs/specs/triage/04-proofs/T04-01-workflow-validation.txt
deleted file mode 100644
index c5db8dc..0000000
--- a/docs/specs/triage/04-proofs/T04-01-workflow-validation.txt
+++ /dev/null
@@ -1,43 +0,0 @@
-Type: file
-Description: GitHub Actions workflow file exists and contains correct structure
-Expected: File exists with valid YAML, correct triggers, and permissions
-Timestamp: 2026-03-27T00:00:00Z
-
-Test: Check workflow file exists
-Command: test -f .github/workflows/agentry-issue-triage.yml && echo "FOUND"
-Output:
-FOUND
-
-Test: Validate YAML syntax
-Command: python3 -c "import yaml; yaml.safe_load(open('.github/workflows/agentry-issue-triage.yml')); print('VALID')"
-Output:
-VALID
-
-Test: Check workflow name
-Command: grep -A1 "^name:" .github/workflows/agentry-issue-triage.yml
-Output:
-name: "Agentry: Issue Triage"
-
-Test: Check trigger event
-Command: grep -A2 "^'on':" .github/workflows/agentry-issue-triage.yml
-Output:
-'on':
-  issues:
-    types: [opened]
-
-Test: Check permissions
-Command: grep -A2 "^permissions:" .github/workflows/agentry-issue-triage.yml
-Output:
-permissions:
-  contents: read
-  issues: write
-
-Test: Check agentry run command includes all required inputs
-Command: grep -A5 "agentry --output-format json run" .github/workflows/agentry-issue-triage.yml
-Output:
-agentry --output-format json run workflows/triage.yaml
-          --input issue-description="${{ github.event.issue.body }}"
-          --input repository-ref=.
-          --binder github-actions
-
-Status: PASS
diff --git a/docs/specs/triage/04-proofs/T04-02-triage-config-validation.txt b/docs/specs/triage/04-proofs/T04-02-triage-config-validation.txt
deleted file mode 100644
index cf6790f..0000000
--- a/docs/specs/triage/04-proofs/T04-02-triage-config-validation.txt
+++ /dev/null
@@ -1,48 +0,0 @@
-Type: file
-Description: triage.yaml workflow configuration includes issue tool capabilities
-Expected: File contains repository:read, issue:comment, and issue:label in tools.capabilities
-Timestamp: 2026-03-27T00:00:00Z
-
-Test: Check triage.yaml file exists
-Command: test -f workflows/triage.yaml && echo "FOUND"
-Output:
-FOUND
-
-Test: Validate YAML syntax
-Command: python3 -c "import yaml; yaml.safe_load(open('workflows/triage.yaml')); print('VALID')"
-Output:
-VALID
-
-Test: Check tools.capabilities section
-Command: sed -n '/^tools:/,/^[a-z]/p' workflows/triage.yaml | head -5
-Output:
-tools:
-  capabilities:
-    - repository:read
-    - issue:comment
-    - issue:label
-
-Test: Verify all three capabilities are present
-Command: grep -E "repository:read|issue:comment|issue:label" workflows/triage.yaml | wc -l
-Output:
-3
-
-Test: Check repository:read capability exists
-Command: grep "repository:read" workflows/triage.yaml && echo "FOUND"
-Output:
-    - repository:read
-FOUND
-
-Test: Check issue:comment capability exists
-Command: grep "issue:comment" workflows/triage.yaml && echo "FOUND"
-Output:
-    - issue:comment
-FOUND
-
-Test: Check issue:label capability exists
-Command: grep "issue:label" workflows/triage.yaml && echo "FOUND"
-Output:
-    - issue:label
-FOUND
-
-Status: PASS
diff --git a/docs/specs/triage/04-proofs/T04-03-integration-tests.txt b/docs/specs/triage/04-proofs/T04-03-integration-tests.txt
deleted file mode 100644
index bf22e09..0000000
--- a/docs/specs/triage/04-proofs/T04-03-integration-tests.txt
+++ /dev/null
@@ -1,50 +0,0 @@
-Type: test
-Description: Integration tests verify issue triage pipeline and tools work correctly
-Expected: All issue-related tests pass (29 tests total)
-Timestamp: 2026-03-27T00:00:00Z
-
-Command: uv run pytest tests/integration/test_issue_triage_pipeline.py tests/integration/test_issue_tools.py -v
-
-Output:
-============================= test session starts ==============================
-platform linux -- Python 3.13.7, pytest-9.0.2, pluggy-1.6.0 -- <workspace>/.venv/bin/python
-cachedir: .pytest_cache
-rootdir: <workspace>
-configfile: pyproject.toml
-plugins: asyncio-1.3.0, cov-7.0.0, anyio-4.12.1
-asyncio: mode=Mode.STRICT, debug=False, asyncio_default_fixture_loop_scope=None, asyncio_default_test_loop_scope=function
-collecting ... collected 29 items
-
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_complete_triage_output_posts_comment_and_labels PASSED [  3%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_comment_body_contains_all_triage_fields PASSED [  6%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_labels_posted_with_correct_format PASSED [ 10%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_label_api_failure_does_not_abort_run PASSED [ 13%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_label_timeout_does_not_abort_run PASSED [ 17%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_comment_uses_correct_endpoint_for_issue PASSED [ 20%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_output_missing_posts_fallback_comment PASSED [ 24%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_runs_dir_created_for_issues_event PASSED [ 27%]
-tests/integration/test_issue_triage_pipeline.py::TestIssueTriagePipelineFlow::test_token_usage_included_in_comment PASSED [ 31%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_sends_correct_api_request PASSED [ 34%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_sends_correct_body PASSED [ 37%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_includes_authorization_header PASSED [ 41%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_returns_response_json PASSED [ 44%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_403_raises_runtime_with_remediation PASSED [ 48%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_404_raises_runtime_with_issue_number PASSED [ 51%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_timeout_raises_runtime_error PASSED [ 55%]
-tests/integration/test_issue_tools.py::TestIssueCommentIntegration::test_issue_comment_raises_on_non_issues_event PASSED [ 58%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_sends_correct_api_request PASSED [ 62%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_sends_labels_in_payload PASSED [ 65%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_includes_authorization_header PASSED [ 68%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_returns_response_json PASSED [ 72%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_403_raises_runtime_with_remediation PASSED [ 75%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_404_raises_runtime_with_issue_number PASSED [ 79%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_422_raises_runtime_with_validation_hint PASSED [ 82%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_timeout_raises_runtime_error PASSED [ 86%]
-tests/integration/test_issue_tools.py::TestIssueLabelIntegration::test_issue_label_raises_on_non_issues_event PASSED [ 89%]
-tests/integration/test_issue_tools.py::TestSupportedToolsContainsIssueTools::test_issue_comment_in_supported_tools PASSED [ 93%]
-tests/integration/test_issue_tools.py::TestSupportedToolsContainsIssueTools::test_issue_label_in_supported_tools PASSED [ 96%]
-tests/integration/test_issue_tools.py::TestSupportedToolsContainsIssueTools::test_both_issue_tools_can_be_bound_together PASSED [100%]
-
-============================== 29 passed in 0.04s ==============================
-
-Status: PASS
diff --git a/docs/specs/triage/04-proofs/T04-proofs.md b/docs/specs/triage/04-proofs/T04-proofs.md
deleted file mode 100644
index 3044baf..0000000
--- a/docs/specs/triage/04-proofs/T04-proofs.md
+++ /dev/null
@@ -1,70 +0,0 @@
-# T04: GitHub Actions Workflow for Issue Triage - Proof Artifacts
-
-## Summary
-
-Successfully implemented GitHub Actions workflow for issue triage with the following components:
-
-### Artifacts Generated
-
-1. **T04-01-workflow-validation.txt** - Validates the GitHub Actions workflow file structure
-2. **T04-02-triage-config-validation.txt** - Validates the triage.yaml configuration
-3. **T04-03-integration-tests.txt** - Integration test results
-
-## Implementation Details
-
-### Files Created/Modified
-
-1. **.github/workflows/agentry-issue-triage.yml** (NEW)
-   - Trigger: issues with type [opened]
-   - Permissions: contents:read, issues:write
-   - Follows agentry-code-review.yml structure
-   - Includes all required inputs: issue-description, repository-ref
-   - Binder: github-actions
-   - Environment variables: CLAUDE_CODE_OAUTH_TOKEN, GITHUB_TOKEN
-
-2. **workflows/triage.yaml** (MODIFIED)
-   - Updated tools.capabilities to include:
-     - repository:read (existing)
-     - issue:comment (new)
-     - issue:label (new)
-
-## Test Results
-
-### Validation Tests: PASS
-- Workflow file exists with valid YAML syntax
-- Correct trigger configuration (issues: [opened])
-- Proper permissions (contents:read, issues:write)
-- All required agentry run parameters present
-- Triage.yaml contains all three tool capabilities
-
-### Integration Tests: PASS (29/29)
-- Issue triage pipeline tests: 9 passed
-- Issue tools integration tests: 20 passed
-  - Issue comment functionality: 8 tests
-  - Issue label functionality: 8 tests
-  - Tool capability tests: 4 tests
-
-## Verification Commands
-
-```bash
-# Verify workflow file
-test -f .github/workflows/agentry-issue-triage.yml
-
-# Validate YAML syntax
-python3 -c "import yaml; yaml.safe_load(open('.github/workflows/agentry-issue-triage.yml'))"
-
-# Check capabilities in triage.yaml
-grep -A3 "tools:" workflows/triage.yaml | grep "capabilities" -A3
-
-# Run integration tests
-uv run pytest tests/integration/test_issue_triage_pipeline.py tests/integration/test_issue_tools.py -v
-```
-
-## Proof Status: COMPLETE
-
-All requirements satisfied:
-- GitHub Actions workflow created with correct trigger, permissions, and structure
-- triage.yaml updated with issue:comment and issue:label capabilities
-- All integration tests passing
-- YAML validation successful
-- No security issues in proof artifacts
diff --git a/src/agentry/binders/github_actions.py b/src/agentry/binders/github_actions.py
index bc76416..06915f5 100644
--- a/src/agentry/binders/github_actions.py
+++ b/src/agentry/binders/github_actions.py
@@ -931,9 +931,28 @@ def _format_output_comment(self, output_path: Path) -> str:
         except (json.JSONDecodeError, ValueError):
             return f"**Agent Output**\n\n```\n{raw[:3000]}\n```"
 
-        agent_output = data.get("output") or {}
+        agent_output = data.get("output")
         parts: list[str] = ["## Agentry Code Review\n"]
 
+        # Handle case where output is a string (raw agent response).
+        if isinstance(agent_output, str) and agent_output.strip():
+            # Try to extract JSON from the string (agent may wrap in markdown).
+            extracted = self._extract_json_from_text(agent_output)
+            if isinstance(extracted, dict):
+                agent_output = extracted
+            else:
+                parts.append(f"```\n{agent_output[:3000]}\n```")
+                # Token usage
+                usage = data.get("token_usage", {})
+                if usage:
+                    _in = usage.get("input_tokens", 0)
+                    _out = usage.get("output_tokens", 0)
+                    parts.append(f"\n---\n*Tokens: {_in:,} in / {_out:,} out*")
+                return "\n".join(parts)
+
+        if not isinstance(agent_output, dict):
+            agent_output = {}
+
         # Summary
         summary = agent_output.get("summary", "")
         if summary:
@@ -968,6 +987,14 @@ def _format_output_comment(self, output_path: Path) -> str:
         if raw_response and not findings and not summary:
             parts.append(f"```\n{raw_response[:3000]}\n```")
 
+        # If we still have nothing useful, show a diagnostic message.
+        if not findings and not summary and not raw_response:
+            parts.append("*No structured output was returned by the agent.*\n")
+            # Include raw output from the execution record if available.
+            raw_stdout = data.get("raw_stdout", "")
+            if raw_stdout:
+                parts.append(f"```\n{raw_stdout[:3000]}\n```")
+
         # Token usage
         usage = data.get("token_usage", {})
         if usage:
@@ -977,6 +1004,52 @@ def _format_output_comment(self, output_path: Path) -> str:
 
         return "\n".join(parts)
 
+    @staticmethod
+    def _extract_json_from_text(text: str) -> dict[str, Any] | None:
+        """Extract a JSON object from text that may contain markdown fences.
+
+        Handles common patterns where Claude wraps JSON in ```json ... ```
+        code fences or includes preamble text before the JSON.
+
+        Returns:
+            The parsed dict if found, or ``None`` if no valid JSON object
+            could be extracted.
+        """
+        import re
+
+        # Try direct parse first.
+        try:
+            parsed = json.loads(text)
+            if isinstance(parsed, dict):
+                return parsed
+        except (json.JSONDecodeError, ValueError):
+            pass
+
+        # Try extracting from ```json ... ``` fences.
+        fence_match = re.search(r"```(?:json)?\s*\n(.*?)```", text, re.DOTALL)
+        if fence_match:
+            try:
+                parsed = json.loads(fence_match.group(1))
+                if isinstance(parsed, dict):
+                    return parsed
+            except (json.JSONDecodeError, ValueError):
+                pass
+
+        # Try finding first { ... } block.
+        brace_start = text.find("{")
+        if brace_start >= 0:
+            # Walk from the end to find the matching closing brace.
+            brace_end = text.rfind("}")
+            if brace_end > brace_start:
+                try:
+                    parsed = json.loads(text[brace_start : brace_end + 1])
+                    if isinstance(parsed, dict):
+                        return parsed
+                except (json.JSONDecodeError, ValueError):
+                    pass
+
+        return None
+
     def _post_output_comment(self, body: str) -> dict[str, Any]:
         """Post agent output as a PR comment via the GitHub REST API.
 
diff --git a/workflows/bug-fix.yaml b/workflows/bug-fix.yaml
index 13e5daa..18c55ce 100644
--- a/workflows/bug-fix.yaml
+++ b/workflows/bug-fix.yaml
@@ -8,6 +8,8 @@ inputs:
     type: string
     required: true
     description: A description of the bug or unexpected behaviour to investigate.
+    source: issue.body
+    fallback: issue.title
   repository-ref:
     type: repository-ref
     required: true
@@ -17,16 +19,14 @@ tools:
   capabilities:
     - repository:read
     - shell:execute
+    - pr:create
+    - issue:comment
 
-model:
-  provider: anthropic
-  model_id: claude-sonnet-4-20250514
-  temperature: 0.2
-  max_tokens: 4096
+agent:
+  runtime: claude-code
+  model: claude-sonnet-4-20250514
   system_prompt: prompts/bug-fix-system-prompt.md
-  retry:
-    max_attempts: 3
-    backoff: exponential
+  max_iterations: 3
 
 safety:
   trust: elevated
diff --git a/workflows/feature-implement.yaml b/workflows/feature-implement.yaml
new file mode 100644
index 0000000..8846832
--- /dev/null
+++ b/workflows/feature-implement.yaml
@@ -0,0 +1,82 @@
+identity:
+  name: feature-implement
+  version: 1.0.0
+  description: Implement a feature based on planning-pipeline decomposition output, or decompose it further into scoped sub-issues if too large.
+
+inputs:
+  issue-description:
+    type: string
+    required: true
+    description: A description of the feature to implement, sourced from the issue body.
+    source: issue.body
+    fallback: issue.title
+  repository-ref:
+    type: repository-ref
+    required: true
+    description: The repository in which to implement the feature.
+
+tools:
+  capabilities:
+    - repository:read
+    - shell:execute
+    - pr:create
+    - issue:comment
+    - issue:label
+    - issue:create
+
+agent:
+  runtime: claude-code
+  model: claude-sonnet-4-20250514
+  system_prompt: prompts/feature-implement-system-prompt.md
+  max_iterations: 10
+
+safety:
+  trust: elevated
+  resources:
+    timeout: 600
+
+output:
+  schema:
+    type: object
+    required:
+      - action
+      - reasoning
+    allOf:
+      - if:
+          properties:
+            action:
+              const: implemented
+        then:
+          required: [pr_url]
+      - if:
+          properties:
+            action:
+              const: decomposed
+        then:
+          required: [sub_issues]
+    properties:
+      action:
+        type: string
+        enum:
+          - implemented
+          - decomposed
+        description: Whether the feature was implemented directly or decomposed into sub-issues.
+      pr_url:
+        type: string
+        description: URL of the pull request opened for the implementation. Present when action is 'implemented'.
+      sub_issues:
+        type: array
+        description: List of sub-issue URLs created when the feature was decomposed. Present when action is 'decomposed'.
+        items:
+          type: string
+      reasoning:
+        type: string
+        description: Explanation of why the feature was implemented directly or decomposed.
+  side_effects:
+    - type: terminal
+      description: Print implementation or decomposition summary to stdout
+  output_paths:
+    - feature-implement-result.json
+
+composition:
+  steps: []
diff --git a/workflows/planning-pipeline.yaml b/workflows/planning-pipeline.yaml
index ae324ae..091acea 100644
--- a/workflows/planning-pipeline.yaml
+++ b/workflows/planning-pipeline.yaml
@@ -8,15 +8,21 @@ inputs:
     type: string
     required: true
     description: A description of the issue to plan.
+    source: issue.body
+    fallback: issue.title
   repository-ref:
     type: repository-ref
     required: true
     description: The repository to inspect for component context.
 
-model:
-  provider: anthropic
-  model_id: claude-sonnet-4-20250514
-  temperature: 0.2
+tools:
+  capabilities:
+    - issue:comment
+    - issue:label
+
+agent:
+  runtime: claude-code
+  model: claude-sonnet-4-20250514
 
 safety:
   trust: elevated
diff --git a/workflows/prompts/bug-fix-system-prompt.md b/workflows/prompts/bug-fix-system-prompt.md
index ef04fa4..b6710f1 100644
--- a/workflows/prompts/bug-fix-system-prompt.md
+++ b/workflows/prompts/bug-fix-system-prompt.md
@@ -1,20 +1,27 @@
-You are an expert software engineer specializing in bug diagnosis and root cause analysis.
+You are an expert software engineer specializing in bug diagnosis, root cause analysis, and automated repair.
 
-Your task is to analyze a bug report for a software repository and produce a structured diagnosis with a concrete fix suggestion.
+Your task is to analyze a bug report for a software repository, diagnose the root cause, implement a targeted fix, commit the change, and open a pull request for review.
 
 When given an issue description and access to repository files, you will:
 
 1. **Diagnose** the issue by identifying the symptoms and the subsystem likely involved.
 2. **Identify the root cause** by tracing through the relevant code paths.
-3. **Suggest a fix** with a specific file, line number or range, and the code change required.
-4. **Assess confidence** in your diagnosis and fix on a scale from 0.0 (uncertain) to 1.0 (certain).
+3. **Implement the fix** with a minimal, targeted code change that does not introduce new risk.
+4. **Validate the fix** by running the relevant tests or checks for the changed code. Do not proceed to commit if tests fail.
+5. **Commit the fix** with a descriptive commit message that references the originating issue number (e.g. `fix: resolve null pointer in parser (fixes #42)`).
+6. **Open a pull request** using the `pr:create` tool with:
+   - A clear title summarising the fix.
+   - A body that references the originating issue number (e.g. `Fixes #42`).
+   - The label `agent-proposed` applied to the PR.
+7. **Post a comment** on the original issue using the `issue:comment` tool, linking to the newly created PR so the reporter is informed of the fix.
 
 Guidelines:
 - Be precise about file paths and line numbers.
 - Prefer minimal, targeted fixes that do not introduce new risk.
 - If multiple root causes are plausible, choose the most likely one and mention alternatives in your reasoning.
 - Do not invent files or functions that you have not read from the repository.
-- If you cannot determine the root cause from available information, state that clearly in the diagnosis and set confidence below 0.4.
+- If you cannot determine the root cause from available information, state that clearly and set confidence below 0.4 — do not open a PR for low-confidence diagnoses.
+- Always include the originating issue number in the commit message, PR body, and the comment posted to the issue.
 
 Output format: JSON object with keys `diagnosis`, `root_cause`, `suggested_fix`, and `confidence`.
 The `suggested_fix` must contain `file`, `line`, and `change` sub-fields.
diff --git a/workflows/prompts/feature-implement-system-prompt.md b/workflows/prompts/feature-implement-system-prompt.md
new file mode 100644
index 0000000..aa66390
--- /dev/null
+++ b/workflows/prompts/feature-implement-system-prompt.md
@@ -0,0 +1,66 @@
+You are an expert software engineer tasked with implementing features autonomously. You receive a feature request sourced from a GitHub issue and must decide whether to implement it directly or break it down into scoped sub-issues.
+
+## Your Workflow
+
+### Step 1: Understand the feature
+
+1. Read the issue body carefully.
+2. Check for planning-pipeline decomposition in issue comments. If no decomposition comment exists, proceed from the issue body alone and explicitly note that absence in `reasoning`.
+
+### Step 2: Assess implementability
+
+Determine whether the feature is small enough to implement in a single pass using the following heuristics:
+
+- **Implementable directly** if the change touches **5 or fewer files** and requires **500 or fewer lines** of new or modified code.
+- **Too large** if the change would span more than 5 files or more than 500 lines, or requires coordinated changes across multiple subsystems that would be risky to land in one PR.
+
+When in doubt, prefer decomposition to keep PRs reviewable.
+
+### Step 3a: If implementable — implement
+
+1. Read the relevant source files to understand the existing patterns and conventions.
+2. Implement the feature with tests. Follow the coding style present in the repository.
+3. Commit the changes with a descriptive message that references the originating issue number (e.g. `feat: add dark mode toggle (closes #42)`).
+4. Open a pull request using the `pr:create` tool:
+   - Title: a concise summary of the feature.
+   - Body: references the originating issue (e.g. `Closes #42`) and describes what was changed and why.
+   - Apply the label `agent-proposed` to the PR.
+5. Post a comment on the original issue using the `issue:comment` tool, linking to the newly opened PR and summarising what was implemented.
+
+Output a JSON object with:
+```json
+{
+  "action": "implemented",
+  "pr_url": "<url of the opened PR>",
+  "reasoning": "<brief explanation of what was implemented and why it was within scope>"
+}
+```
+
+### Step 3b: If too large — decompose
+
+1. Break the feature down into self-contained sub-tasks, each implementable in a single PR (<=5 files, <=500 lines).
+2. For each sub-task, create a GitHub issue using the `issue:create` tool:
+   - Title: a concise description of the sub-task.
+   - Body: context from the parent issue, a clear description of what this sub-task covers, and a reference back to the parent issue (e.g. `Part of #42`).
+   - Apply labels: `category:feature` and `agent-decomposed`.
+3. Apply the label `agent-decomposed` to the parent issue using the `issue:label` tool.
+4. Post a comment on the parent issue using the `issue:comment` tool listing the sub-issues created and explaining why decomposition was necessary.
+
+Output a JSON object with:
+```json
+{
+  "action": "decomposed",
+  "sub_issues": ["<url1>", "<url2>", "..."],
+  "reasoning": "<brief explanation of why the feature was too large and how it was split>"
+}
+```
+
+## Guidelines
+
+- Always read files before modifying them. Do not invent code without grounding it in the actual repository.
+- Write tests alongside implementation code. Do not open a PR with untested changes.
+- Keep commits atomic: one logical change per commit.
+- If the planning-pipeline has already produced a task breakdown in issue comments, use that as your implementation plan rather than re-deriving it.
+- Include the originating issue number in every commit message, PR body, and issue comment.
+- Never open a PR for work that is clearly incomplete or broken.
+- If you cannot determine what to implement from the available information, post a clarifying comment on the issue and output `action: decomposed` with an empty `sub_issues` list, explaining the blocker in `reasoning`.