[getappz]

Summary

Fixes "redispatch to a new agent doesn't stick" (seen on agentflare item #604, where a redispatch to claude-code kept reverting to opencode).

Root cause. agentflare_backend::item::release set assignee_agent = NULL whenever it released a claim, whoever the assignee was. If a job for agent A is still running when the operator redispatches to agent B, that job later fails, release_and_comment -> item_release -> item::release wipes B. The queue then retries the same job, apply_fresh_overrides sees a NULL assignee and fails open to the job's frozen payload agent (A), and the claim re-assigns the item to A:<job>.

Evidence from item_assignment_events on #604:

time	from	to
12:04:20	opencode:job	claude-code (redispatch)
12:05:54	None	opencode:job (retry's claim)

The NULL between those rows is the wipe.

Fix. item::release now clears assignee_agent only when the item has not been reassigned to a different agent type than the releasing owner (agent_part comparison). Ordinary release still clears it, so item #93's behavior (a released item must not stay pinned to the releasing agent type) is unchanged, and a no-op release is unchanged too. This complements #780, which releases the old claim on redispatch but could not stop a still-running job from wiping the assignee afterwards.

Not changed: apply_fresh_overrides fail-open (a NULL assignee after a normal release legitimately means "use the payload agent"), redispatch, reassignment_releases_claim.

Tracks agentflare item #605 (not GitHub PR/issue #605).

Test plan

• Both new tests failed before the fix with left: None, right: Some("claude-code"), and pass after
  • release_keeps_an_assignee_reassigned_to_a_different_agent_mid_run (backend)
  • item_release_by_a_superseded_job_keeps_the_redispatched_assignee (MCP layer, replays redispatch then the old job's release)
• cargo test -p agentflare-backend: 151 passed
• cargo fmt --all --check clean
• cargo clippy --locked --workspace --all-targets --all-features -- -D warnings -A unsafe_code -A clippy::pedantic exits 0
• Full cargo test --workspace: not clean locally, see below; left to CI

Local cargo test --bin agentflare had 5 failures, none caused by this change. Two fail identically on unmodified master (agent_detector_flags_the_claudecode_marker_run_denies_on, which depends on the session env, and claim_self_repair_defers_when_another_workstation_already_holds_a_live_claim); pm_health_matches_item_health_for_the_same_params is a 1-second timing flake; the other two (claim_guard_leaves_the_claim_held_when_disarmed, claim_then_headless_not_found_releases_claim_and_posts_error_comment) pass when run alone.

Known follow-ups (not in this PR)

• A redispatch does not stop the still-running old job; it keeps executing until it fails.
• Unconfirmed: pick_implementer_agent may swap claude-code for a fallback when claude_over_threshold() is true. No evidence it fired here.

Summary by CodeRabbit

• Bug Fixes

  • Preserved the current assignee when a previous agent releases an item after it has been reassigned.
  • Prevented outdated releases from clearing newer assignments.

• Tests

  • Added coverage for reassignment scenarios across item handling and dispatch workflows.

[getappz]

Tracking this PR for automated review (flared:51bb8de6c33b).

[coderabbitai]

Understand this PR’s impact

Explore downstream dependencies and potential security impact with Blast Radius.

View blast radius →

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository: getappz/agentflare/.coderabbit.yaml

Review profile: CHILL

Plan: Essentials

Run ID: e828e2c0-2971-4142-bebd-0328b8b52ad2

📥 Commits

Reviewing files that changed from the base of the PR and between f153897 and 554582a.

📒 Files selected for processing (3)

• crates/agentflare-backend/src/item/claim.rs
• crates/agentflare-backend/src/item/tests.rs
• src/mcp_server/tests/action_tests.rs

Included review availability: 1 review is currently available. Your included PR review attempts over the past 7 days set your current allowance at 2 reviews per hour.

---

📝 Walkthrough

Walkthrough

release now preserves an assignee when another agent reassigns the item before the original claim releases. Backend and MCP tests cover this behavior.

Changes

Lease release safety

Layer / File(s)	Summary
Preserve reassigned ownership and validate release behavior crates/agentflare-backend/src/item/claim.rs, crates/agentflare-backend/src/item/tests.rs, src/mcp_server/tests/action_tests.rs	release checks for reassignment by canonical agent type before clearing the assignee. Regression tests verify that claude-code remains assigned after an obsolete opencode:1 claim releases.

Priority: ⬇️ Low

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly identifies the primary change: preventing item release from erasing a reassignment made during an active run.
Description check	✅ Passed	The description follows the required Summary and Test plan sections, explains the root cause and fix, documents regression tests, and reports validation results and known follow-ups. The template's No…
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 7 functions across 3 files.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

🧪 Generate unit tests (beta)

• Commit to this branch
• Create a new PR

---

_{Comment @coderabbitai help to get the list of available commands.}

[getappz]

supervisor — CI self-repair dispatched

Failing checks: build (windows-latest), CI Green.

Please investigate and push a fix.

job: 5jRhX0yJWtupNBYK_jS0V

[getappz]

supervisor — CI green

Checks are passing again.