Skip to content

fix(review): preserve long sessions in .github/workflows/opencode-review.yml #564

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
seonghobae wants to merge 9 commits into
base: main
Choose a base branch
from
Open

fix(review): preserve long sessions in .github/workflows/opencode-review.yml #564

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
2a2b6f8
fix(review): bootstrap explicit probe evidence
seonghobae Jul 15, 2026
49b9e2e
fix(review): preserve long-running model results
seonghobae Jul 15, 2026
f44c4be
test(review): cover trusted binding boundaries
seonghobae Jul 15, 2026
f206cce
fix(review): preserve long runs and fail over outages
seonghobae Jul 15, 2026
2a45022
chore(review): split receipt normalization follow-up
seonghobae Jul 15, 2026
255a7b7
fix(review): bind probe evidence to literal path lines
seonghobae Jul 15, 2026
d8f1b89
fix(review): provide verified bootstrap receipts
seonghobae Jul 15, 2026
6004eba
fix(review): inline complete verified probe evidence
seonghobae Jul 15, 2026
bd6598d
test(review): seal bootstrap receipt digests
seonghobae Jul 15, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
44 changes: 24 additions & 20 deletions .github/workflows/opencode-review.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1801,6 +1801,7 @@ jobs:
ContextualWisdomLab/.github:tests/test_javascript_coverage_gate.py | \
ContextualWisdomLab/.github:tests/test_opencode_agent_contract.py | \
ContextualWisdomLab/.github:tests/test_opencode_model_pool_runner.py | \
ContextualWisdomLab/.github:tests/test_opencode_review_normalize_output.py | \
ContextualWisdomLab/.github:tests/test_pr_head_replay_guard.py | \
ContextualWisdomLab/.github:tests/test_pr_review_fix_scheduler_coverage.py | \
ContextualWisdomLab/.github:tests/test_pr_review_merge_scheduler.py | \
Expand Down Expand Up @@ -3368,7 +3369,7 @@ jobs:
- name: Run OpenCode PR Review model pool
id: opencode_review_model_pool
if: needs.coverage-evidence.result == 'success'
timeout-minutes: 65
timeout-minutes: 205
continue-on-error: true
env:
STRIX_GITHUB_MODELS_TOKEN: ${{ secrets.STRIX_GITHUB_MODELS_TOKEN || github.token }}
Expand All @@ -3395,13 +3396,16 @@ jobs:
# the SAME model 5x let a rate-limited/hung leader consume the whole
# step, so the pool never reached a healthy fallback model.
OPENCODE_MODEL_ATTEMPTS: "1"
# Keep stale providers from pinning required review jobs for hours.
# Adversarial validation needs enough room to read the evidence, but
# dynamic cadence and the outer watchdog still bound each current-head run.
OPENCODE_RUN_TIMEOUT_SECONDS: "600"
OPENCODE_EXPORT_TIMEOUT_SECONDS: "120"
OPENCODE_TOTAL_RETRY_BUDGET_SECONDS: "3600"
OPENCODE_POOL_STEP_TIMEOUT_SECONDS: "2100"
# Preserve reviews that legitimately need tens of minutes to inspect a
# large repository. Small changes get 30 minutes per model; medium,
# large, unknown, and central-governance changes can use 90 minutes so
# repositories that need a full hour are not cut off at the boundary.
# The 195-minute pool budget leaves room for a second substantial
# candidate, while provider 5xx errors fail over immediately.
OPENCODE_RUN_TIMEOUT_SECONDS: "5400"
OPENCODE_EXPORT_TIMEOUT_SECONDS: "180"
OPENCODE_TOTAL_RETRY_BUDGET_SECONDS: "11700"
OPENCODE_POOL_STEP_TIMEOUT_SECONDS: "12000"
# Keep cycling through the high-sensitivity candidate catalog until
# the retry budget or step timeout is exhausted; a single invalid
# cycle can be all provider formatting noise rather than review
Expand All @@ -3410,16 +3414,16 @@ jobs:
OPENCODE_DYNAMIC_REVIEW_CADENCE: "true"
OPENCODE_SMALL_CHANGE_FILE_THRESHOLD: "3"
OPENCODE_MEDIUM_CHANGE_FILE_THRESHOLD: "20"
OPENCODE_SMALL_CHANGE_RUN_TIMEOUT_SECONDS: "600"
OPENCODE_SMALL_CHANGE_TOTAL_BUDGET_SECONDS: "3600"
OPENCODE_MEDIUM_CHANGE_RUN_TIMEOUT_SECONDS: "600"
OPENCODE_MEDIUM_CHANGE_TOTAL_BUDGET_SECONDS: "3600"
OPENCODE_LARGE_CHANGE_RUN_TIMEOUT_SECONDS: "600"
OPENCODE_LARGE_CHANGE_TOTAL_BUDGET_SECONDS: "3600"
OPENCODE_UNKNOWN_CHANGE_RUN_TIMEOUT_SECONDS: "600"
OPENCODE_UNKNOWN_CHANGE_TOTAL_BUDGET_SECONDS: "3600"
OPENCODE_DYNAMIC_RUN_TIMEOUT_CAP_SECONDS: "600"
OPENCODE_DYNAMIC_TOTAL_BUDGET_CAP_SECONDS: "1800"
OPENCODE_SMALL_CHANGE_RUN_TIMEOUT_SECONDS: "1800"
OPENCODE_SMALL_CHANGE_TOTAL_BUDGET_SECONDS: "4200"
OPENCODE_MEDIUM_CHANGE_RUN_TIMEOUT_SECONDS: "3600"
OPENCODE_MEDIUM_CHANGE_TOTAL_BUDGET_SECONDS: "7800"
OPENCODE_LARGE_CHANGE_RUN_TIMEOUT_SECONDS: "5400"
OPENCODE_LARGE_CHANGE_TOTAL_BUDGET_SECONDS: "11700"
OPENCODE_UNKNOWN_CHANGE_RUN_TIMEOUT_SECONDS: "5400"
OPENCODE_UNKNOWN_CHANGE_TOTAL_BUDGET_SECONDS: "11700"
OPENCODE_DYNAMIC_RUN_TIMEOUT_CAP_SECONDS: "5400"
OPENCODE_DYNAMIC_TOTAL_BUDGET_CAP_SECONDS: "11700"
OPENCODE_DYNAMIC_MAX_CYCLES_CAP: "0"
# This installation currently reports a 4k request-body limit for
# GitHub Models GPT-5 endpoints even though the public catalog is
Expand All @@ -3429,8 +3433,8 @@ jobs:
OPENCODE_DYNAMIC_MAX_CYCLES: "0"
CENTRAL_REVIEW_PROCESS_FALLBACK_ELIGIBLE: ${{ steps.central_review_process_fallback_scope.outputs.eligible || 'false' }}
CENTRAL_REVIEW_PROCESS_FALLBACK_SCOPE_LABEL: ${{ steps.central_review_process_fallback_scope.outputs.scope_label || 'unsupported' }}
OPENCODE_CENTRAL_REVIEW_PROCESS_FALLBACK_RUN_TIMEOUT_SECONDS: "600"
OPENCODE_CENTRAL_REVIEW_PROCESS_FALLBACK_TOTAL_BUDGET_SECONDS: "3600"
OPENCODE_CENTRAL_REVIEW_PROCESS_FALLBACK_RUN_TIMEOUT_SECONDS: "5400"
OPENCODE_CENTRAL_REVIEW_PROCESS_FALLBACK_TOTAL_BUDGET_SECONDS: "11700"
OPENCODE_CENTRAL_REVIEW_PROCESS_FALLBACK_MAX_CYCLES: "1"
OPENCODE_BACKOFF_INITIAL_SECONDS: "30"
OPENCODE_BACKOFF_MAX_SECONDS: "30"
Expand Down
23 changes: 23 additions & 0 deletions ci-review-prompt.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -109,6 +109,29 @@ current-head line bytes without its line ending, and whether the hypothesis was
material code/workflow/config/package/test changes and one for non-code changes;
REQUEST_CHANGES needs a confirmed probe anchored to a published finding.

The `evidence` string itself must literally repeat that probe's declared path and
positive line as `path:line`; the separate `path` and `line` fields do not satisfy
this binding. The same string must also name the independent proof anchor actually
used (for example source trace, test/assertion, executed command, log/check/SARIF,
diff, or CodeGraph) and an observed-result word such as confirmed, rejected,
returned, passed, or failed.

For this central-review bootstrap revision, do not guess a cryptographic receipt.
The trusted current-head source provides these two verified source-line anchors;
copy one receipt exactly only after confirming the cited source trace, and remember
that the receipt is a binding rather than proof by itself:

- `Source trace at scripts/ci/run_opencode_review_model_pool.sh:95 confirmed
the dynamic timeout cap defaults to 5400 seconds;
source-line-sha256=8e8933a12547802d7bc9ff7cd99915808de9d4f27559d83783a6506d3ff94118`.
- `Source trace at scripts/ci/run_opencode_review_model_pool.sh:287 confirmed
a detected provider outage is classified as service-unavailable;
source-line-sha256=4f6509bb2883da6b849c2eb5949e4844ec209e5eaa7ab6b60396b4c2fd065e31`.

Evidence using either anchor must still state what source trace was checked and
what it confirmed or rejected. These bootstrap anchors may be removed once the
trusted normalizer derives receipts directly from validated `path:line` fields.

Execution provenance is mandatory. Never claim that React DevTools, Chrome
DevTools, browser DevTools, Playwright, Cypress, or Selenium ran, passed,
confirmed, verified, or observed behavior unless bounded evidence contains a
Expand Down
Loading
Loading