ship-feature
Skill9 starsChanged 38 days agoMITView on GitHub
Autonomously pick the next pending item from docs/BACKLOG.md, research the best approach, plan, implement, test end-to-end through the MCP server with a real agent, and open a PR. Use when asked to "ship the next feature", "work the backlog", or for the daily autonomous feature run. Optional argument - a backlog ID (e.g. B-007) to work on instead of the top item.
- Commits and pushes
16 other projects use a version of this.
Adopted by
Most recently taken up by
- FAHTHIZ/oasistheir own version, first seen Nov 16, 2026
- theexperiencecompany/gaiatheir own version, first seen Nov 13, 2026
- seanmcgary/claude-skillstheir own version, first seen Nov 11, 2026
- babs/skillstheir own version, first seen Nov 8, 2026
- azents/azentstheir own version, first seen Oct 23, 2026
- mergewatch/mergewatch.aitheir own version, first seen Oct 18, 2026
Other versions16 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 16
- classroomio/classroomio★ 1.7ksame name
- hoangsonww/Claude-Code-Agent-Monitor★ 1.2ksame name
- ultimo-rs/ultimo★ 802same name
- AmineDjeghri/awesome-os-setup★ 673same name
- theexperiencecompany/gaia★ 327same name
- seldonframe/seldonframe★ 40same name
- mergewatch/mergewatch.ai★ 26same name
- azents/azents★ 16same name
- babs/skills★ 4same name
- FAHTHIZ/oasis★ 4same name
6 more not listed
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 repositoryOne other file it gives its agents.
- fix-bugSkill
The file
We split this skill into the rules it gives an agent
11 rules. 0 are common, 2 are distinctive, and 9 are very specific to this project.
0
Common
Most of the 16 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
9
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 23
If the working tree is dirty, stop and report — don't touch someone else's work.
- Very specific: line 27
Read docs/BACKLOG.md. Candidates have Status: pending.
- Very specific: line 28
Remove items already being worked: gh pr list --state open --label autonomous --json title,headRefName — skip any ID that appears in an …
- Very specific: line 30
If $ARGUMENTS names an ID, take that one.
- Distinctive: line 50
Write the plan into your notes (it goes in the PR body later): files to change, public tool signatures, how method selection stays
- Very specific: line 64
Update docs/tools/*.md, the tool table in README.md, and scripts/install.py if install behaviour changes.
- Very specific: line 66
Add pytest tests in tests/ (happy path, declined/edge path, determinism — same input twice gives same method and output).
- Very specific: line 72
Protocol: write a steps file into a temp dir and run uv run python scripts/e2e/mcp_call.py --steps <file> (see its docstring.
- Very specific: line 84
If something fails: debug, fix, re-run all three.
- Very specific: line 100
Commit (conventional message, e.g.
- Very specific: line 109
Put the PR number into the backlog PR: line, commit, push.
“Common” is measured: kept by at least 6 in 10 of the 16 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
Ship one backlog item, end to end, without a human
You are the daily feature agent for meelu-analytics-mcp. Finish one backlog
item completely: research → plan → review the plan → implement → test (unit +
MCP protocol + real agent) → review the code → PR. Never ask the user
questions; make the call yourself and write down why in the PR. Never merge,
never push to main, never force-push.
0. Preflight
git fetch origin && git checkout main && git pull --ff-only
gh label create autonomous --color 5319e7 --description "Opened by an autonomous agent" 2>/dev/null || true
uv sync --extra dev --extra insights
uv run pytest -q # note any failures that already exist on mainIf the working tree is dirty, stop and report — don't touch someone else's work.
1. Pick the item
- Read
docs/BACKLOG.md. Candidates have**Status:** pending. - Remove items already being worked:
gh pr list --state open --label autonomous --json title,headRefName— skip any ID that appears in an open PR's title or branch. - If
$ARGUMENTSnames an ID, take that one. Otherwise take the first candidate by order (the file is sorted by priority). Prefer items withEffort: SorM. If the top item isL, split it: ship a coherent first slice now and add the rest to the backlog as new items. - Branch:
git checkout -b auto/<id-lowercase>-<short-slug>.
2. Research (do this properly — it decides the quality of the result)
Launch subagents in parallel, in one message:
- Web research (general-purpose): current best libraries/approaches for the item as of today — maintenance status, license (must permit commercial use; flag anything else), install weight, Python 3.10–3.12 + macOS/Linux wheels, determinism. Ask for cited sources and a recommendation.
- Codebase exploration (Explore): where the change belongs (read
docs/architecture.md), the closest existing tool to copy patterns from, how results/trust/declined are built (src/tabint/shared/results.py,src/tabint/shared/honesty.py), and the docs page underdocs/tools/to update.
3. Plan, then get it reviewed
Write the plan into your notes (it goes in the PR body later):
files to change, public tool signatures, how method selection stays
deterministic and recorded in metadata, how the trust block and
declined path behave, new deps (optional extras + lazy import, like
insights), tests, and the exact end-to-end checks you'll run.
Then launch a Plan subagent to critique it against the codebase and the two invariants. Address its points before writing code.
4. Implement
- Match the surrounding code: naming, comment density, idioms, result shapes.
- Heavy or niche deps go in an optional extra and are imported lazily; a missing extra must produce an honest declined result, not a traceback.
- Update
docs/tools/*.md, the tool table inREADME.md, andscripts/install.pyif install behaviour changes. - Add pytest tests in
tests/(happy path, declined/edge path, determinism — same input twice gives same method and output).
5. Verify end to end (all three must pass)
- Unit:
uv run pytest -q— no new failures versus preflight. - Protocol: write a steps file into a temp dir and run
uv run python scripts/e2e/mcp_call.py --steps <file>(see its docstring;"$session_key"is substituted for you). Use a realistic generated CSV (write it with pandas/numpy in a temp dir) and include an edge case that should decline. Check: nois_error,metadatarecords the method, atrustblock exists, numbers match an independent pandas/scipy computation. - Real agent:
scripts/e2e/agent_e2e.sh "<plain-English question a user would ask that needs the new feature, with the absolute CSV path>". Confirm the agent called the new/changed tool and the answer is correct and conveys trust. If theclaudeCLI isn't available (some cloud runners), say so in the PR and treat step 2 as the protocol-level proof instead.
If something fails: debug, fix, re-run all three. If after a serious effort it
still can't be made to work, open the PR as draft, explain the blocker
precisely, and set the backlog item to blocked.
6. Independent code review
Launch a general-purpose subagent with only: the backlog item text, git diff
main...HEAD, and the instruction to find correctness bugs, invariant
violations (non-deterministic selection, missing/dishonest trust), missing
tests, and style drift. Fix every real finding; re-run step 5 if code changed.
7. Update the backlog and ship
- In
docs/BACKLOG.mdset the item to**Status:** in-reviewand append- **PR:** <filled after creation>; add any follow-ups you discovered as newpendingitems with the next free ID, in priority position. - Commit (conventional message, e.g.
feat(<area>): ...) ending with:Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> git push -u origin HEADgh pr create --base main --label autonomous --title "<ID>: <title>" --body ...Body sections: Backlog item, Research (recommendation + sources), Design / decisions (including anything you decided without asking), Verification (pytest summary, protocol output excerpt, agent question + tool calls + answer excerpt), Limitations / follow-ups. End the body with:🤖 Generated with [Claude Code](https://claude.com/claude-code)- Put the PR number into the backlog
PR:line, commit, push.
8. Report
Finish with: item ID and title, PR URL, what shipped, how it was verified, and anything a human should look at before merging.
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.

