shipit
Skill2 starsChanged yesterdayMITView on GitHub
End-to-end feature delivery pipeline - grill requirements, run specify→plan→tasks→implement with model split, review/fix, then optionally PR/CI and merge/release.
5 other projects use a version of this.
Adopted by
Most recently taken up by
- nazmulnahid-git/Ai-Stacktheir own version, first seen Oct 26, 2026
- freelawproject/courtlistenertheir own version, first seen Oct 8, 2026
- fire17/shipittheir own version, first seen Jul 6, 2026
- AmosChenZixuan/Agentic-working-contracttheir own version, first seen Jul 2, 2026
- palpen/claude-skillstheir own version, first seen Apr 14, 2026
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- freelawproject/courtlistener★ 1.1ksame name
- nazmulnahid-git/Ai-Stack★ 4same name
- AmosChenZixuan/Agentic-working-contract★ 2same name
- fire17/shipit★ 1same name
- palpen/claude-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository17 other files it gives its agents.
- autopilotSkill
- autopsySkill
- confidence-mapSkill
- delegateSkill
- expert-askSkill
- failure-museumSkill
- ghost-usersSkill
- grill-dumpSkill
- intent-diffSkill
- md-to-mapSkill
- mind-mapSkill
- premortemSkill
- red-teamSkill
- stuck-detectorSkill
- taste-thiefSkill
- tournamentSkill
- workSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
26 rules. 0 are common, 19 are distinctive, and 7 are very specific to this project.
0
Common
Most of the 5 edited versions we read keep these.
19
Distinctive
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 8
Run a complete feature-delivery workflow from a feature description to optional PR/merge/release.
- Distinctive: line 10
Not for small fixes or one-file changes: just make the change.
- Very specific: line 17
If Spec Kit is not installed, stop and say so, naming the missing commands and the link above.
- Distinctive: line 21
Never skip a failed step silently.
- Distinctive: line 22
Never exceed the fix-cycle caps (5 for CI, 5 for release, 2 for review).
- Distinctive: line 23
Never merge or release unless Phase 1 locked it.
- Distinctive: line 27
If the user invoked this skill with a feature description, use it.
- Very specific: line 28
If no description was provided, ask in normal assistant chat for the description and wait.
- Very specific: line 29
After a description is available, grill the requirements first.
- Very specific: line 33
Use AskUserQuestion to collect and confirm all decisions up front.
- Distinctive: line 47
If PR path is selected, post-CI action (collect now, do not ask later):
- Very specific: line 61
After Phase 1 completes, do not call AskUserQuestion again.
- Distinctive: line 68
Ensure implementation includes spec artifacts in commits/PR by default.
- Very specific: line 70
Run expensive review with claude-opus-5-5 (code-review agent).
- Distinctive: line 73
Stop when no material findings remain, or after 2 review/fix cycles.
- Distinctive: line 74
If locked decision is “stop after implementation”, stop here and report outcomes.
- Distinctive: line 78
If locked decision is PR path:
- Distinctive: line 83
Wait for PR pipelines.
- Distinctive: line 84
If pipelines fail, fix and rerun until green or max 5 CI fix cycles reached.
- Distinctive: line 85
If locked PR outcome is “stop when green”, stop and report.
- Distinctive: line 89
If locked PR outcome includes merge+release:
- Distinctive: line 93
Wait for release pipeline.
- Distinctive: line 94
If release pipeline fails, fix and retry until success or max 5 release fix cycles reached.
- Distinctive: line 107
Keep updates concise and outcome-first.
- Distinctive: line 108
Do not silently skip failed steps.
- Distinctive: line 109
If max retry cap is reached, stop with a clear blocker summary and exact failing pipeline/run links.
“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
shipit
Run a complete feature-delivery workflow from a feature description to optional PR/merge/release.
Not for small fixes or one-file changes: just make the change. Use this when the feature merits a spec, plan and task list.
Requirements (check first)
- Spec Kit installed in the project, so the slash commands
/speckit-specify,/speckit-plan,/speckit-tasksand/speckit-implementexist (github.com/github/spec-kit). - Optional: a
grillingskill and acode-reviewagent. Withoutgrilling, run the requirements interview yourself (step 3). Withoutcode-review, use a general-purpose agent with a review brief. git, andghor the repo host's CLI if the PR path is chosen.
If Spec Kit is not installed, stop and say so, naming the missing commands and the link above. Do not improvise the specify/plan/tasks/implement pipeline by hand, and do not run any phase.
Hard rules
- No
AskUserQuestionafter Phase 1. - Never skip a failed step silently.
- Never exceed the fix-cycle caps (5 for CI, 5 for release, 2 for review).
- Never merge or release unless Phase 1 locked it.
Invocation contract
- If the user invoked this skill with a feature description, use it.
- If no description was provided, ask in normal assistant chat for the description and wait. Do not use
AskUserQuestionfor this step. - After a description is available, grill the requirements first. If a
grillingskill is installed, invoke it. Otherwise interview the user yourself: ask the open decisions in rounds, give a recommended answer to each, and wait for replies before moving on.
Phase 1: Grill-and-lock session (interactive)
Use AskUserQuestion to collect and confirm all decisions up front. This is the only interactive phase.
You must confirm each run:
- The feature description/scope (using the provided description as starting point).
- Model split:
/speckit-specify=claude-opus-5-5(default, expensive)/speckit-plan=claude-opus-5-5(default, expensive)/speckit-tasks=claude-opus-5-5(default, expensive)/speckit-implement=claude-sonnet-5-5(default, cheaper)
- Optional invocation overrides (if provided in user text), but still confirm before execution.
- Post-implementation action (collect now, do not ask later):
- Recommended: create PR, wait for CI, auto-fix failures
- Alternative: stop after implementation + review recommendations are applied
- If PR path is selected, post-CI action (collect now, do not ask later):
- Recommended: stop when PR pipelines are green
- Alternative: merge using default merge options, then cut a release
- Release bump policy (if release path might be used):
- Suggest patch for bugfix/compat-only work
- Suggest minor for new user-visible capability
- Confirm the selected bump during this initial session
- Failure-handling cap:
- Up to 5 auto-fix cycles for CI/release failures
At the end of this phase, produce a compact locked execution summary and proceed autonomously.
Phase 2: Autonomous execution (no more AskUserQuestion)
After Phase 1 completes, do not call AskUserQuestion again.
- Run:
/speckit-specify(claude-opus-5-5 by default)/speckit-plan(claude-opus-5-5 by default)/speckit-tasks(claude-opus-5-5 by default)/speckit-implement(claude-sonnet-5-5 by default)
- Ensure implementation includes spec artifacts in commits/PR by default.
- Review/fix loop:
- Run expensive review with claude-opus-5-5 (
code-reviewagent). - Apply fixes with claude-sonnet-5-5.
- Re-review with claude-opus-5-5.
- Stop when no material findings remain, or after 2 review/fix cycles.
- Run expensive review with claude-opus-5-5 (
- If locked decision is “stop after implementation”, stop here and report outcomes.
Phase 3: PR/CI (conditional)
If locked decision is PR path:
- Commit changes.
- Push branch.
- Create PR.
- Wait for PR pipelines.
- If pipelines fail, fix and rerun until green or max 5 CI fix cycles reached.
- If locked PR outcome is “stop when green”, stop and report.
Phase 4: Merge + release (conditional)
If locked PR outcome includes merge+release:
- Merge PR using default merge options.
- Cut release with the previously confirmed bump type (patch/minor).
- Wait for release pipeline.
- If release pipeline fails, fix and retry until success or max 5 release fix cycles reached.
Final report
End with: feature, branch/PR link, per-phase result, review cycles used, and release tag if any. Five lines or fewer, then TL;DR:.
Anti-patterns
- Asking the user something new mid-run instead of using the locked summary.
- Reading the setup's model names as fixed: confirm them in Phase 1, since they are defaults.
- Calling the spec commands before Spec Kit is confirmed present.
- Leaving spec artifacts out of the commit.
Operational rules
- Keep updates concise and outcome-first.
- Do not silently skip failed steps.
- If max retry cap is reached, stop with a clear blocker summary and exact failing pipeline/run links.
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.

