itsryanlenk/humanist / plugins/humanist/skills/ship-gate
ship-gate
VerifiedThe release gate for public prose. Runs every quality gate in a fixed order before a draft ships, and refuses to call a draft ready while any gate is unrun. Use before publishing or handing over anything a reader will see: a blog post, a web page, a newsletter, a product listing, release notes, a social post, a client deliverable. Trigger phrases include "is this ready to ship", "run the gates", "final check before publishing", "ship gate", "red team this", "check this before it goes live". Orchestrates the humanist rewrite pass, the humanist.py checker, the composition read and an adversary pass that attacks the claims, then writes a gate report. It does not edit the draft on its own authority and it does not publish anything.
Ship gate: no draft is ready until every gate has a result
The humanist skill makes a draft read better. This skill decides whether it may
ship. The two jobs are different. A draft can read well and still carry a wrong
number, a stale figure or a claim nobody checked.
The one rule. A gate that did not run is a failed gate. A missing tool, a broken shell, a deadline or a long session never turns "not run" into "pass".
Why this exists
Every gate in a prose pipeline is a step that someone has to choose to do. Under pressure the gates get skipped, and the skip gets reported as a tooling problem or not reported at all. The draft then reaches the reader with the one defect the skipped gate would have caught.
This skill makes a skip visible. Each gate must end with one of three results, and the draft carries its status in its own name until all of them pass.
The gates, in order
Run them in this order. Each one assumes the ones before it ran.
| # | Gate | What it catches | How it runs |
|---|---|---|---|
| 1 | Rewrite pass | Machine residue, filler, inflated significance | ../humanist/SKILL.md step 1 |
| 2 | Checker | Mechanical tells, banned constructions, dashes | ../humanist/humanist.py |
| 3 | Composition read | Rhythm tells no pattern catches | ../humanist/SKILL.md step 3 |
| 4 | Adversary, lens A | Wrong claims, missing bases, weak sources | A subagent with adversary.md |
| 5 | Adversary, lens B | What lens A missed, and defects its fixes added | A second, fresh subagent |
| 6 | Re-measure | Anything the fixes from gates 4 and 5 broke | Gate 2 again, on the final text |
| 7 | Render check | Copy that is correct in the file and wrong on the page | Open the published surface or its preview |
Gate 7 applies when the draft goes into a CMS, a template or any system that changes what the reader sees. A plain file handed over as-is skips it, and the report says so.
Gate 2 details
python ../humanist/humanist.py draft.md --strip-quotes
Ship at zero FAIL. Judge each WARN. For web copy, extract the body first with
../humanist/tools/html2prose.py, as the humanist skill describes. Exit code 2
means the checker did not run. Record it as NOT RUN, never as a pass.
Gates 4 and 5: the adversary
The adversary is a subagent. It receives the draft, the primary sources the draft
cites, and adversary.md. It does not receive the drafting conversation.
An adversary that knows why each choice was made will defend those choices. That is the failure this gate exists to prevent.
Run it twice, with fresh context each time and a different lens each time:
- Lens A, the skeptic. Runs every pass in
adversary.md. - Lens B, the fact checker. Verifies every number, date, quote and named source against the primary sources, one by one.
The second run is not redundant. Fixes made after the first run add new defects: a length limit breached by an added sentence, or a new claim with no source. The second run and the re-measure in gate 6 exist to catch those.
Verify a blocker yourself before you act on it. A subagent can invent a mechanism too.
When a tool fails
Each gate names a job. The usual tool is one way to do that job. When the tool fails, do the same job another way.
| Failure | What to do |
|---|---|
| The checker cannot run where the draft lives | Copy the draft and the checker to a place where Python runs, and run it there. |
| No subagent is available | Start a fresh session that holds only the draft, the sources and adversary.md. |
| The preview cannot render | Mark gate 7 NOT RUN and say so in the report. Do not describe the page as checked. |
| Nothing works | Stop. Report the gate as NOT RUN with the exact error text. Do not ship. |
Report the exact error message, not a summary of it. A summary hides whether the failure was real and whether a fallback existed.
Time pressure
When time runs short, cut scope. Never cut gates.
Ship fewer changes with every gate passed, or stage all of them marked NOT READY with the owed gates listed. Both are acceptable. A draft marked ready with gates unrun is not.
The status label
Until every gate passes, the draft carries NOT READY in the place a reviewer will see first: the file name, the CMS title or the first line of the handoff. The person approving it should not have to read a report to learn that it failed.
Remove the label only when the gate report shows every gate as PASS or as N/A with a reason.
The gate report
End every run with this block. Put it at the top of any handoff.
SHIP GATE: READY | NOT READY
Draft: <file or URL>
Run at: <date and time, from a checked clock>
| # | Gate | Result | Evidence |
|---|-----------------|---------|---------------------------------------|
| 1 | Rewrite pass | PASS | <what changed, in one line> |
| 2 | Checker | PASS | 0 FAIL, 3 WARN judged |
| 3 | Composition | PASS | <what the read changed> |
| 4 | Adversary A | FIX | 2 blockers, both fixed |
| 5 | Adversary B | PASS | 0 blockers |
| 6 | Re-measure | PASS | 0 FAIL after fixes |
| 7 | Render check | NOT RUN | <exact error text> |
Owed before ship: <every NOT RUN gate, with its fallback>
Allowed results: PASS, FAIL, FIX (blockers found and fixed), N/A (with a reason) and NOT RUN (with the exact error). A report with any FAIL or NOT RUN is NOT READY.
What this skill must never do
- Never report a draft as ready while any gate is NOT RUN.
- Never let a deadline decide which gates run.
- Never give the adversary the drafting history.
- Never edit claims on its own authority. Blockers go to the author, who decides.
- Never publish. The person who owns the surface publishes.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |