Ship
Drive a direct candidate, lift slice, or lift promotion through the right evidence and PR lifecycle. Runs exactly one official review for a direct/final candidate, consumes delta receipts for lift promotion, distinguishes technical readiness from human merge authority, and only queues/watches a merge when explicitly authorized. Invoke as `$ship` / `/ship`, or when asked to finish, land, merge, make a PR mergeable, or make it pristine. Works identically under Codex and Claude Code.From its SKILL.md
npx -y skills add comment-hq/skills --skill shipAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
6.8 KB, ~1.3k tokens by cl100k_base, as published. Nobody here has run it
ship — certify and move the intended delivery boundary
Read delivery-methodology, the repo's AGENTS.md/CLAUDE.md, and linked
delivery/testing/merge docs first. Determine the actual PR base and choose one
mode before running gates.
Modes
Direct candidate
A complete independently shippable task branch targeting the release branch. Freeze one commit, run the complete affected lane and one official review on that SHA, then push/open the PR.
Lift slice
A bounded delta targeting a declared controlled lift. Require focused checks and a base/head review receipt. Push/open the slice PR and merge it into the lift with a real merge commit when the human-approved lift plan authorizes internal slice merges. It does not enter the release merge queue or rerun the cumulative main-based affected lane by default.
Lift promotion
The frozen lift targeting the release branch. Require: no open dependent slice; current release base merged into the lift; conflict/uncovered delta receipted; ordered receipt coverage verified; complete affected lane plus required integration/migration evidence on the frozen SHA; and one composition-oriented official review.
Exactly one official review path
For a direct candidate or lift promotion, prefer local codex review; otherwise
use the runtime's PR-based code-review, never both. Run the affected lane and
local official review concurrently when practical. Unknown/partial results are
not passes.
A PR-based-only reviewer may require opening a draft after test certification. Keep it draft until that same SHA passes review. A finding or edit returns only the affected delta to convergence and creates a new final candidate; valid unrelated lift receipts remain valid.
Worklog and identity
The worklog is optional for standalone use. When a caller supplies a worklog or
Project Root, keep its route identity/token and update only material transitions:
mode/frozen SHA, receipt, technically ready, blocker, queued, and merged. Never
switch to an ambient registered or Botlets profile. Invoke comment-identity
only immediately before an uncredentialed direct-REST write.
Workflow
- Converge locally. Finish the bounded implementation/review batch and focused checks. Record invariants and residual risk.
- Freeze. Commit intended state, require a clean worktree, and record SHA, intended base, and mode.
- Certify the mode:
- lift slice → focused evidence + accepted delta receipt;
- direct → complete affected lane + one official candidate review;
- promotion → receipt-chain verification + complete affected/integration evidence + one official composition review.
- Push the exact SHA. Use a normal explicit-SHA refspec; reserve
--force-with-leasefor an intentional verified rewrite (never rewrite a controlled lift). Verify remote head, then create/update the PR. Use a draft only for the PR-based-review exception above. - Report technical readiness. Re-confirm the current head, required PR checks, mergeability, and unresolved review threads. PR-event jobs that skip are not evidence; rely on the local certification. Address one complete late in-scope finding batch, then recertify the affected candidate. An unrelated discovery becomes a focused GitHub issue and does not expand this PR; stop only if it makes the candidate unsafe, then ask the owner human for a separate worktree job.
- Do not wait indefinitely for an external reaction. If repository rules or the human require approval and it is absent, report “technically ready, awaiting approval” and end/resume on an actual event. Never manufacture a reaction or keep the session alive without a real monitor.
- Merge only with authority:
- lift slice → merge sequentially into its lift with a real merge commit when the approved lift plan authorizes it; record merge SHA in the receipt;
- direct/promotion → only when the user explicitly asked to merge/land/queue, enqueue through the repository's release-branch mechanism with a real merge commit and watch until actually merged.
- Verify ancestry. Fetch the actual base and prove the reviewed/frozen SHA landed. After promotion, prove every recorded slice head and frozen lift head are ancestors of the release branch. Clean up only after verification.
Technical-ready criteria
- Remote head equals the certified local SHA.
- Required evidence for the chosen mode is current.
- Required PR checks are passing/skipped as designed and mergeability is not dirty.
- No known actionable blocker or unresolved required thread remains.
- Residual risks/declines are recorded.
Technical readiness is distinct from human merge authority and actual merge.
Guardrails
- Never push directly to the release branch, squash-merge, or deploy production without explicit authority.
- If the user said no push/PR, stop after local evidence and report.
- Preserve stricter domain gates for production deployment, incidents, migrations, or protocol cut-over when they protect irreversible risk.
- Do not delay a user-useful candidate for speculative enterprise hardening, abstraction, or unsupported edge cases. Ship the simplest boundary that satisfies acceptance and hard invariants.
Repo config
Run this repo's commands. Read AGENTS.md (else CLAUDE.md) and linked delivery/testing/merge/deploy docs. If absent, infer focused and final lanes from package/build/CI config and offer comment-init.
Comment.io API
Use the active worklog/root and its working Comment.io route first. Resolve and freeze $BASE for the whole workflow in this order: the supplied comm's validated final Comment.io origin after any shortlink redirect; the active Comment.io tool/account base URL; an explicitly selected profile's base_url; only when no target context exists, https://comment.io. A shortlink origin is never $BASE; do not switch a staging/custom workflow to production. For direct REST, consult $BASE/llms/reference.txt only when exact API or recovery detail is needed. Fetch $BASE/llms.txt only when no current route works or another focused guide is needed. Invoke comment-identity only before an uncredentialed direct-REST write; never replace a supplied token or tool/browser/connector identity. Don't restate the live contracts here.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.
Gives 0 of the 12 instructions most ship operate skills give in ~1.3k tokens
Counted across 779 of the 1,178 authors here whose files we hold, read 2026-08-07
- Document a rollback plan before deploymentin 41 of 779, across 22 files
- Update the changelogin 21 of 779, across 19 files
- Run the test suitein 20 of 779
- Create an annotated git tagin 20 of 779
- Clean up feature flags after full rolloutin 18 of 779, across 10 files
- Verify deployment health after launchin 18 of 779, across 10 files
- Test both feature flag statesin 17 of 779, across 9 files
- Verify the working tree is cleanin 17 of 779
- Make database migrations backward-compatiblein 16 of 779, across 8 files
- Set up error monitoring before launchin 15 of 779, across 7 files
- Monitor metrics at each rollout stagein 14 of 779, across 5 files
- Create a GitHub releasein 14 of 779
Said here and by no other author read
- read delivery-methodology and repo configuration docs first
- freeze exactly one commit before running gates
- run the complete affected test lane
- run exactly one official review path
- push the exact frozen commit SHA
- report technical readiness status
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.