issue-workbench
Skill1 starsChanged 4 months agoApache-2.0View on GitHub
Implement one GitHub issue in a guarded feature branch. Use when asked to build or fix one issue, run review-checkpoint, publish a PR, or return a Shipyard integration handoff path.
In this repository19 other files it gives its agents.
- AGENTS.mdAGENTS.md
- agent-aeoSkill
- agent-memorySkill
- beanquerySkill
- ci-repairbaySkill
- delegated-developmentSkill
- ghSkill
- issue-blueprintSkill
- minifluxSkill
- og-imageSkill
- opSkill
- pencil-to-figma-migrationSkill
- pr-comment-sweepSkill
- pr-launchpadSkill
- repo-surveyorSkill
- review-checkpointSkill
- review-repairbaySkill
- shipyardSkill
- skill-optimizerSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 4 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 26
Default lite mode handles one issue only.
- Distinctive: line 27
Stop on unclear acceptance or product behavior, unrelated dirty changes, forbidden-path changes not explicitly required, or required …
- Distinctive: line 30
Do not add compatibility, migration, aliases, fallbacks, dependencies, or future-proofing without an explicit requirement.
- Distinctive: line 41
If truncation hides scope, rerun with larger limits.
- Distinctive: line 62
Use git add -N for new files.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Issue Workbench
Implement one actionable issue with the smallest intentional diff. $review-checkpoint owns blocker actionability; $pr-launchpad owns PR publication; Shipyard owns integration.
Inputs
issue_numberis required.base_branchdefaults to the repository default;branch_slugis optional.worktree_pathis optional in pull-request mode and is prepared by Shipyard in integration mode.max_iterationsandreview_basepass through to$review-checkpoint.handoff_modeispull_request(default) orintegration_branch.integration_branchand Shipyard's preparedworktree_pathare required in integration mode.
Memory
Direct invocation owns issue-scoped $agent-memory; integration mode defers it
to Shipyard.
Boundaries
- Default lite mode handles one issue only; do not invoke Blueprint, Shipyard, parent/child planning, or
final_check. - Stop on unclear acceptance or product behavior, unrelated dirty changes, forbidden-path changes not explicitly required, or required unrelated refactoring.
- Modify only issue-owned files. Do not touch secrets, env, generated, lock,
.agents/, infrastructure, or.context/files unless explicitly required or a verified review blocker requires it. .context/may contain only local progress, review, memory, decision, and handoff artifacts. Keep progress togoal,current_step,artifacts,blockers, andvalidation; never commit it.- Do not add compatibility, migration, aliases, fallbacks, dependencies, or future-proofing without an explicit requirement. Use Conventional Commits and stage only inspected paths.
- Integration children send no interim status, never push their local branches, and never run
pr-launchpad.
Workflow
In integration mode, enter Shipyard's existing
worktree_path; require its current branch to matchissue-<issue_number>(-<slug>)?, requireintegration_branchto be its ancestor, and never create or switch the child branch. Then require a clean worktree and read the issue with:python3 <skill_dir>/scripts/issue_snapshot.py <issue_number>If truncation hides scope, rerun with larger limits. Extract requirements and constraints; do not invent behavior.
In pull-request mode, prepare the branch with:
python3 <skill_dir>/scripts/start_issue_branch.py <issue_number> [--base-branch <base_branch>] [--branch-slug <branch_slug>] [--worktree-path <worktree_path>]review_basedefaults tointegration_branchin integration mode andorigin/<base_branch>otherwise. Preserve any initialized five-field progress file.- Inspect relevant code, callers, existing patterns, and tests. Select the public behavior seam and smallest meaningful validation; do not add implementation-detail tests when no useful seam exists. Implement only the issue.
Inspect the complete diff and run the path guard, which includes untracked files:
git status --short git diff --stat git diff python3 <skill_dir>/scripts/diff_guard.py --base <review_base>Use
git add -Nfor new files. Allow a blocked path only after tracing it to an explicit issue requirement or verified review blocker. Run the smallest relevant validation, commit inspected paths, and bind its command and result tovalidated_head=$(git rev-parse HEAD).- Capture
pre_review_head, then run$review-checkpointwithmode=fix_loop,review_base, andmax_iterations; do not pass Shipyard's shared manifest from a child worktree. ReturnBLOCKEDwith its findings instead of treating it as PASS. Continue only after the latest completed review returnsPASSwith no later commit. IfHEADchanged, rerundiff_guard.pyand use the review fix's check evidence for the newvalidated_head; otherwise reuse Step 4 evidence. - Finish according to mode.
Handoff
In pull-request mode, after PASS, invoke nested $pr-launchpad with
validated_head and its validation
commands/results, then return only the PR URL.
In integration mode, pass inspected facts to the helper; it writes the canonical .context/integration-handoff.json through Shipyard's manifest interface:
# PASS
python3 <skill_dir>/scripts/integration_child.py finish --review-base <review_base> --review PASS --check "<cmd>" --known-skip "<reason>"
# defect owned by another child
python3 <skill_dir>/scripts/integration_child.py finish --review-base <review_base> --review FAIL --needs-child-fix '#123'Return only the helper's absolute path; never reconstruct, copy, or extend its payload or choose a Shipyard child status.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

