[codex] fix agent-team manual workflow_dispatch inputs#69
Merged
Conversation
github-actions Bot
added a commit
that referenced
this pull request
Apr 23, 2026
…dening PR #69 added an explicit required-input contract to the planner, implementer, and reviewer agents, and hardened the implementer's optional pr_number handling. Update catalog/agent-team/README.md to reflect the exact fail-loud protocol, the two-branch recovery path (input-propagation failure vs. generic block), and the pr_number unresolved-literal treatment. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
3 tasks
github-actions Bot
added a commit
that referenced
this pull request
Apr 24, 2026
… section PR #69 added loud-failure guards when workflow_dispatch inputs are missing or unresolved. Users now need concrete gh workflow run commands to re-dispatch a stuck stage — the previous docs mentioned manual dispatch was possible but gave no commands. Add explicit commands for each role and a Troubleshooting section covering the three most common stuck-pipeline scenarios. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 tasks
github-actions Bot
added a commit
that referenced
this pull request
Apr 25, 2026
Two documentation fixes surfaced by the #69 dispatch-inputs hardening: 1. README.md uninstall section: `state:in-progress` → `agent-team:reviewed`. The install skill, E2E tests, and reviewer workflow all use `agent-team:reviewed` as the 7th label. `state:in-progress` never existed in the pipeline. 2. catalog/agent-team/README.md: expand "Retrying a blocked task" to cover mid-pipeline re-dispatch after a `workflow_dispatch inputs were not propagated` block (the new fail-loud behavior added in #69). Old text only described restarting from spec. New text adds explicit `gh workflow run` commands for planner/ implementer/reviewer with all required inputs. Also updates the "Input propagation" gotcha entry to explain the error, label it expected behavior (not a bug), and cross-reference the recovery steps. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This was referenced Apr 25, 2026
github-actions Bot
added a commit
that referenced
this pull request
Apr 27, 2026
The "Input propagation" bullet previously said only "must fail loudly" without describing the actual behavior introduced in #69. Expand it to show the two-path failure logic (issue_number present vs missing), the exact error message, the recovery command, and a separate bullet explaining the implementer's optional pr_number semantics. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
github-actions Bot
added a commit
that referenced
this pull request
Apr 28, 2026
…examples
- README.md uninstall section listed `state:in-progress` which doesn't
exist; the correct label created by the install skill is
`agent-team:reviewed`. Fixes the stale label name.
- catalog/agent-team/README.md step 4 now includes concrete
`gh workflow run` commands for re-dispatching each agent role manually.
Each workflow reads inputs via `${{ github.event.inputs.* }}` (fixed in
PR #69); the examples make the correct `-f` flags explicit so users can
retry a stuck stage without guessing the input names.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
github-actions Bot
added a commit
that referenced
this pull request
May 1, 2026
PR #69 hardened all three dispatch-receiving agents (planner, implementer, reviewer) to fail loudly when workflow_dispatch inputs are missing or arrive as unresolved template literals. The catalog README was updated in that PR, but the install-agent-team SKILL.md still listed escape hatches without mentioning the input contract. Add a "Manual re-dispatch" bullet to the step-8 install summary and extend the user-journey escape-hatch line to include the required --field flags and the fail-loud behaviour, so users reading only the skill file get the full picture before attempting a manual gh workflow run. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
github-actions Bot
added a commit
that referenced
this pull request
May 4, 2026
Add a per-role input reference table to the "Kicking off a task" section so users know exactly which inputs are required vs optional when manually re-dispatching planner, implementer, or reviewer. Document the exact blocked-state error message and how to recover from a `state:blocked` caused by missing dispatch inputs (as opposed to a human-added block), since the recovery path differs. Expand the "Input propagation" gotcha to include the failure message and cross-reference the new table. Follows the behavioral fix in #69 (fail loud on missing workflow_dispatch inputs, harden optional implementer pr_number). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
github-actions Bot
added a commit
that referenced
this pull request
May 5, 2026
- catalog/agent-team/README.md: add `pr_number` lifecycle section to the handoff model (blank on first impl → creates PR; reviewer passes it back on kickback → implementer pushes to existing branch). Expand "Input propagation" gotcha with the exact error message, unresolved-literal detection, and the optional-by-design note for pr_number. - CONTRIBUTING.md: replace "no automated test harness" with the actual tiered test strategy (tier-2 invariants, tier-1 skill tests, tier-3 E2E) and clear guidance on when to run each tier. Follows PR #69 (fix agent-team dispatch-input hardening). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This was referenced May 5, 2026
github-actions Bot
added a commit
that referenced
this pull request
May 8, 2026
…overy Expand the "Input propagation" gotcha bullet to describe the exact failure mode introduced by PR #69: agents check for unresolved ${{ github.event.inputs.* }} literals (not just empty strings), post a specific 🛑 error comment when inputs are missing, and surface the recovery path. Add a new Troubleshooting section with concrete gh workflow run commands for manually re-dispatching a stalled planner, implementer, or reviewer, and a note explaining how the implementer distinguishes a first-attempt blank pr_number from a kickback real PR number. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This was referenced May 8, 2026
github-actions Bot
added a commit
that referenced
this pull request
May 11, 2026
- README.md: fix incorrect 7th label in uninstall section (`state:in-progress` → `agent-team:reviewed`); expand to explicit per-label `gh label delete` commands for clarity - catalog/agent-team/README.md: expand "Input propagation" gotcha to document the fail-loud behavior added in #69 — unresolved template literal detection, issue_number-missing vs. present handling, and the pr_number blank/literal treatment for first-dispatch vs. kickback - CONTRIBUTING.md: fix "seven dogfooded workflows" → six (three agentic + three CI), and clarify that only the agentic workflows are gh-aw managed Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This was referenced May 11, 2026
github-actions Bot
added a commit
that referenced
this pull request
May 12, 2026
…umber hardening Mirrors the fixes from #69 (fail loud on missing dispatch inputs, harden optional implementer pr_number) in the user-facing README: - Expand the "Input propagation" gotcha to describe the two failure modes (state:blocked comment when issue_number is present; missing_data / report_incomplete when it is absent) and call out that unresolved template literals are treated the same as missing values. - Add a "pr_number handling" gotcha documenting that blank *or* unresolved literal → new branch, not an update to an existing PR. Also fix a broken link in CONTRIBUTING.md: the "Publishing" section it pointed to was removed in v0.2 and never replaced. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
3 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #55
What changed
This fixes the agent-team manual
workflow_dispatchpath so downstream prompts read dispatch inputs using gh-aw's documented markdown expression form instead of relying on bareinputs.*prose.The planner, implementer, and reviewer prompts now fail loudly when required dispatch inputs are missing instead of inferring the target issue from labels or recent activity. The implementer prompt was also hardened so an optional
pr_numberthat arrives blank or as a raw${{ github.event.inputs.pr_number }}literal is treated as not set on first-pass runs.Root cause
Manual dispatches were succeeding at the workflow layer, but the runtime prompt body was not consuming
workflow_dispatchinputs in the documented form. That left the implementer and reviewer vulnerable to missing or unresolved values, and older runs only appeared to work because the agent could infer context from repository state.Impact
Manual redispatch for planner, implementer, and reviewer is now deterministic. If dispatch inputs are missing, the run stops with an explicit blocked state instead of guessing. If
pr_numberis intentionally omitted on a first implementer pass, the prompt now clearly tells the agent to create a new branch/PR rather than treating the unresolved template token as meaningful input.Validation
./tests/test-invariants.shverkyyi/agent-team-playgroundon April 23, 2026verkyyi/agent-team-playgroundon April 23, 2026verkyyi/agent-team-playgroundon April 23, 2026pr_numberhandling guidanceNotes
Playground verification used temporary workflow installation and canary issues/PRs, and the playground repo was restored to baseline afterward.