diff --git a/crates/skilllite-assistant/src-tauri/src/skilllite_bridge/integrations/evolution_ui/authorize.rs b/crates/skilllite-assistant/src-tauri/src/skilllite_bridge/integrations/evolution_ui/authorize.rs
index dd76838..e4ba89c 100644
--- a/crates/skilllite-assistant/src-tauri/src/skilllite_bridge/integrations/evolution_ui/authorize.rs
+++ b/crates/skilllite-assistant/src-tauri/src/skilllite_bridge/integrations/evolution_ui/authorize.rs
@@ -2,9 +2,12 @@
 
 use crate::skilllite_bridge::evolution_cli::spawn_skilllite_json;
 use crate::skilllite_bridge::local::engine_types::AuthorizeCapabilityResponse;
-use crate::skilllite_bridge::local::env_keys::evolution as evo_keys;
 use crate::skilllite_bridge::paths::{find_project_root, load_dotenv_for_child};
 
+fn authorized_evolution_run_args(proposal_id: &str) -> Vec<&str> {
+    vec!["evolution", "run", "--json", "--proposal-id", proposal_id]
+}
+
 pub fn authorize_capability_evolution(
     workspace: &str,
     tool_name: &str,
@@ -38,17 +41,33 @@ pub fn authorize_capability_evolution(
         let root = find_project_root(&workspace_owned);
         let mut cmd = std::process::Command::new(&skilllite_path_owned);
         crate::windows_spawn::hide_child_console(&mut cmd);
-        cmd.arg("evolution")
-            .arg("run")
-            .arg("--json")
+        cmd.args(authorized_evolution_run_args(&proposal_id_owned))
             .current_dir(&root)
             .stdout(std::process::Stdio::piped())
             .stderr(std::process::Stdio::piped());
         for (k, v) in load_dotenv_for_child(&workspace_owned) {
             cmd.env(k, v);
         }
-        cmd.env(evo_keys::SKILLLITE_EVO_FORCE_PROPOSAL_ID, &proposal_id_owned);
         let _ = cmd.output();
     });
     Ok(proposal_id)
 }
+
+#[cfg(test)]
+mod tests {
+    use super::authorized_evolution_run_args;
+
+    #[test]
+    fn authorize_background_run_args_force_proposal() {
+        assert_eq!(
+            authorized_evolution_run_args("proposal_20260604_110000.000"),
+            vec![
+                "evolution",
+                "run",
+                "--json",
+                "--proposal-id",
+                "proposal_20260604_110000.000"
+            ]
+        );
+    }
+}
diff --git a/tasks/TASK-2026-067-authorized-evolution-proposal-run/CONTEXT.md b/tasks/TASK-2026-067-authorized-evolution-proposal-run/CONTEXT.md
new file mode 100644
index 0000000..cdd5f50
--- /dev/null
+++ b/tasks/TASK-2026-067-authorized-evolution-proposal-run/CONTEXT.md
@@ -0,0 +1,28 @@
+# Technical Context
+
+## Current State
+
+- Relevant crates/files: `crates/skilllite-assistant/src-tauri/src/skilllite_bridge/integrations/evolution_ui/authorize.rs`, `crates/skilllite-commands/src/evolution.rs`, `crates/skilllite-evolution/src/run.rs`.
+- Current behavior: `authorize.rs` enqueues a proposal, spawns `skilllite evolution run --json`, sets `SKILLLITE_EVO_FORCE_PROPOSAL_ID`, and discards output. `cmd_run` removes that env var when no `--proposal-id` argument is supplied, so `run_evolution_inner` does not load the authorized proposal.
+
+## Architecture Fit
+
+- Layer boundaries involved: desktop assistant UI bridge -> SkillLite CLI -> command crate -> evolution engine.
+- Interfaces to preserve: Tauri command shape, `skilllite evolution authorize-capability`, `skilllite evolution run --proposal-id`, and workspace dotenv merging.
+
+## Dependency and Compatibility
+
+- New dependencies: none.
+- Backward compatibility notes: existing callers that set `--proposal-id` are unchanged; this fix aligns the authorization background caller with the supported CLI argument.
+
+## Design Decisions
+
+- Decision: pass `--proposal-id` in the background subprocess argument list and keep env propagation harmless.
+  - Rationale: `cmd_run` treats the CLI argument as the authoritative way to force a proposal and scopes the environment variable safely around execution.
+  - Alternatives considered: modify `cmd_run` to preserve a pre-existing env var when no CLI argument is passed.
+  - Why rejected: preserving ambient env would broaden the force semantics for all CLI callers and risk unexpected forced runs from inherited process environments.
+
+## Open Questions
+
+- [x] Is a docs update required? No, this restores the existing documented command behavior and does not change user-facing command syntax.
+- [x] Is a database migration required? No.
diff --git a/tasks/TASK-2026-067-authorized-evolution-proposal-run/PRD.md b/tasks/TASK-2026-067-authorized-evolution-proposal-run/PRD.md
new file mode 100644
index 0000000..121fdf7
--- /dev/null
+++ b/tasks/TASK-2026-067-authorized-evolution-proposal-run/PRD.md
@@ -0,0 +1,37 @@
+# PRD
+
+## Background
+
+Recent desktop split work moved authorized capability evolution to a CLI subprocess. The CLI supports forcing a backlog proposal with `--proposal-id`, and `cmd_run` intentionally scopes the force environment variable around that explicit argument. The desktop authorization path did not pass the argument, so the authorized proposal id could be lost before `run_evolution` selected work.
+
+## Objective
+
+When a user starts evolution from the chat capability prompt, the immediately spawned background run must target the exact proposal id returned by `authorize-capability`.
+
+## Functional Requirements
+
+- FR-1: The authorization background process must include `--proposal-id` followed by the authorized proposal id.
+- FR-2: Existing `.env` and child environment merging must continue to work.
+- FR-3: Manual evolution triggering remains unchanged.
+
+## Non-Functional Requirements
+
+- Security: Do not bypass coordinator policy runtime or auto-approve dangerous evolution; only preserve target selection.
+- Performance: No additional process or database work beyond the existing background run.
+- Compatibility: Preserve the existing `skilllite evolution run --proposal-id` CLI contract and current desktop API shape.
+
+## Constraints
+
+- Technical: Keep the fix local to the desktop bridge unless runtime evidence shows the CLI contract itself is broken.
+- Timeline: N/A for automation execution; complete within this investigation branch.
+
+## Success Metrics
+
+- Metric: authorization background run argument list.
+- Baseline: `skilllite evolution run --json` with only `SKILLLITE_EVO_FORCE_PROPOSAL_ID`.
+- Target: `skilllite evolution run --json --proposal-id <id>`, with optional environment also harmless.
+
+## Rollout
+
+- Rollout plan: ship as a focused bug-fix PR.
+- Rollback plan: revert the assistant bridge argument helper if unexpected subprocess behavior appears.
diff --git a/tasks/TASK-2026-067-authorized-evolution-proposal-run/REVIEW.md b/tasks/TASK-2026-067-authorized-evolution-proposal-run/REVIEW.md
new file mode 100644
index 0000000..6478376
--- /dev/null
+++ b/tasks/TASK-2026-067-authorized-evolution-proposal-run/REVIEW.md
@@ -0,0 +1,29 @@
+# Review Report
+
+## Scope Reviewed
+
+- Files/modules: desktop authorized evolution bridge, `skilllite evolution run` proposal forcing, task artifacts.
+- Commits/changes: recent P2 desktop/CLI split path and this bug-fix branch.
+
+## Findings
+
+- Critical: None.
+- Major: Active bug fixed. `skilllite_authorize_capability_evolution` returned a proposal id to the UI, then spawned a background run without `--proposal-id`. `cmd_run` removed `SKILLLITE_EVO_FORCE_PROPOSAL_ID` when no CLI proposal id was present, so the authorized backlog proposal was not forced.
+- Minor: Assistant crate still emits pre-existing warnings during plain `cargo test` / `cargo clippy`; they are unrelated to this change.
+
+## Quality Gates
+
+- Architecture boundary checks: `pass`
+- Security invariants: `pass`
+- Required tests executed: `pass`
+- Docs sync (EN/ZH): `pass`
+
+## Test Evidence
+
+- Commands run: `cargo fmt --check`; `cargo test --manifest-path crates/skilllite-assistant/src-tauri/Cargo.toml authorize_background_run_args_force_proposal`; `cargo test --manifest-path crates/skilllite-assistant/src-tauri/Cargo.toml`; `cargo clippy --manifest-path crates/skilllite-assistant/src-tauri/Cargo.toml --all-targets`; `cargo test -p skilllite-commands`; `cargo clippy --all-targets -- -D warnings`; `cargo test`; `python3 scripts/validate_tasks.py`.
+- Key outputs: targeted assistant regression test: `1 passed`; full assistant tests: `51 passed`; `skilllite-commands`: `23 passed`; workspace clippy finished successfully with `-D warnings`; workspace test suite passed; task validation passed for 67 task directories.
+
+## Decision
+
+- Merge readiness: `ready`
+- Follow-up actions: None for this fix. Existing assistant warnings can be handled separately.
diff --git a/tasks/TASK-2026-067-authorized-evolution-proposal-run/STATUS.md b/tasks/TASK-2026-067-authorized-evolution-proposal-run/STATUS.md
new file mode 100644
index 0000000..b168b8e
--- /dev/null
+++ b/tasks/TASK-2026-067-authorized-evolution-proposal-run/STATUS.md
@@ -0,0 +1,21 @@
+# Status Journal
+
+## Timeline
+
+- 2026-06-04:
+  - Progress: Identified that desktop authorized capability evolution spawns `skilllite evolution run --json` without `--proposal-id`, while `cmd_run` clears the force env var when the CLI argument is absent. Implemented a minimal fix so the background run passes `--proposal-id <authorized id>` and added `authorize_background_run_args_force_proposal`.
+  - Blockers: None.
+  - Next step: Done after validation.
+- 2026-06-04:
+  - Progress: Validation passed: `cargo fmt --check`; targeted assistant regression test; full assistant tests; assistant clippy; `cargo test -p skilllite-commands`; workspace clippy with `-D warnings`; workspace `cargo test`; task validation.
+  - Blockers: None.
+  - Next step: PR ready.
+
+## Checkpoints
+
+- [x] PRD drafted before implementation (or `N/A` recorded)
+- [x] Context drafted before implementation (or `N/A` recorded)
+- [x] Implementation complete
+- [x] Tests passed
+- [x] Review complete
+- [x] Board updated
diff --git a/tasks/TASK-2026-067-authorized-evolution-proposal-run/TASK.md b/tasks/TASK-2026-067-authorized-evolution-proposal-run/TASK.md
new file mode 100644
index 0000000..a891f09
--- /dev/null
+++ b/tasks/TASK-2026-067-authorized-evolution-proposal-run/TASK.md
@@ -0,0 +1,51 @@
+# TASK Card
+
+## Metadata
+
+- Task ID: `TASK-2026-067`
+- Title: Fix authorized evolution proposal run binding
+- Status: `done`
+- Priority: `P1`
+- Owner: `agent`
+- Contributors: `automation`
+- Created: `2026-06-04`
+- Target milestone: recent critical bug investigation
+
+## Problem
+
+Desktop chat authorization for capability evolution enqueues a concrete backlog proposal and then starts a background `skilllite evolution run`. The background run only sets `SKILLLITE_EVO_FORCE_PROPOSAL_ID`, but `cmd_run` clears that environment variable whenever no `--proposal-id` CLI argument is present. As a result, the user-authorized proposal may remain queued while the background run sees no forced proposal and either does nothing or selects a different candidate.
+
+## Scope
+
+- In scope: bind the authorized background evolution run to the returned proposal id using the existing `--proposal-id` CLI contract.
+- In scope: add a focused regression test for the background argument construction.
+- Out of scope: changing evolution policy runtime decisions, backlog schema, or LLM synthesis behavior.
+
+## Acceptance Criteria
+
+- [x] Background runs started by `skilllite_authorize_capability_evolution` pass `--proposal-id <authorized id>`.
+- [x] The forced proposal environment variable is no longer the only binding mechanism for this path.
+- [x] Tests cover the argument contract and the affected crates still pass required checks.
+
+## Risks
+
+- Risk: argument construction diverges from manual trigger behavior again.
+  - Impact: authorized proposals can be skipped or unrelated proposals can execute.
+  - Mitigation: centralize the authorization background run args in a small helper with a regression test.
+
+## Validation Plan
+
+- Required tests: targeted Rust tests for the assistant bridge and workspace Rust tests.
+- Commands run: `cargo fmt --check`; `cargo test --manifest-path crates/skilllite-assistant/src-tauri/Cargo.toml authorize_background_run_args_force_proposal`; `cargo test --manifest-path crates/skilllite-assistant/src-tauri/Cargo.toml`; `cargo clippy --manifest-path crates/skilllite-assistant/src-tauri/Cargo.toml --all-targets`; `cargo test -p skilllite-commands`; `cargo clippy --all-targets -- -D warnings`; `cargo test`; `python3 scripts/validate_tasks.py`.
+- Manual checks: inspected the final `authorize.rs` command chain and confirmed the child receives `--proposal-id`.
+
+## Regression Scope
+
+- Areas likely affected: desktop chat "start evolution" action, desktop evolution CLI bridge, `skilllite evolution run` proposal forcing.
+- Explicit non-goals: policy runtime thresholds, proposal recovery, queue selection, and UI copy.
+
+## Links
+
+- Source TODO section: N/A, triggered by automated recent-commit bug investigation.
+- Related PRs/issues: recent P2 desktop/CLI split commits.
+- Related docs: N/A, this restores documented CLI behavior without changing user-facing commands.
diff --git a/tasks/board.md b/tasks/board.md
index 43428f7..6380bea 100644
--- a/tasks/board.md
+++ b/tasks/board.md
@@ -1,6 +1,6 @@
 # Task Board
 
-Last updated: 2026-05-29 (TASK-2026-066 utf8 evolution log truncate done)
+Last updated: 2026-06-04 (TASK-2026-067 authorized evolution proposal run done)
 
 ## In Progress
 
@@ -17,6 +17,7 @@ Last updated: 2026-05-29 (TASK-2026-066 utf8 evolution log truncate done)
 
 ## Done
 
+- `TASK-2026-067-authorized-evolution-proposal-run` - Status: `done` - Owner: `agent`
 - `TASK-2026-066-utf8-evolution-log-truncate` - Status: `done` - Owner: `agent`
 - `TASK-2026-064-env-keys-single-source` - Status: `done` - Owner: `agent`
 - `TASK-2026-063-extension-tool-metadata-dispatch` - Status: `done` - Owner: `agent`