publishing-prds
/publishing-prds
Renderuje PRD repozytorium do jednego pliku HTML dla zespołu, biznesu lub recenzenta, bez dopisywania twierdzeń spoza dokumentu.
- Autor
- PRDAnvil ↗
- Repozytorium
- prd-toolkit ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 11.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 11.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Renderuje PRD repozytorium do jednego pliku HTML dla zespołu, biznesu lub recenzenta, bez dopisywania twierdzeń spoza dokumentu.
Wyzwalacze z opisu
Oryginalny opis (EN)
Use when asked to publish, share, export or render the PRD of a repo that has a docs/prd/ folder as a page, an HTML file or an Artifact for a team, the business or a reviewer - "publish the PRD", "make a page of the PRD for the team", /prd-publish - and when asked to put on that page a claim the PRD does not make, error logs, evidence or a secret; prd_publish.py renders the PRD's own text into one self-contained HTML file outside the PRD folder, Claude adds per-area summaries that cite their IDs, and the file is offered as a private Artifact or its path given
Jak działa
- Overview
- Non-negotiable rules
- Publishing, step by step
- Red Flags - STOP
- Common Rationalizations
- Related skills
Podgląd SKILL.md
---
name: publishing-prds
description: Use when asked to publish, share, export or render the PRD of a repo that has a docs/prd/ folder as a page, an HTML file or an Artifact for a team, the business or a reviewer - "publish the PRD", "make a page of the PRD for the team", /prd-publish - and when asked to put on that page a claim the PRD does not make, error logs, evidence or a secret; prd_publish.py renders the PRD's own text into one self-contained HTML file outside the PRD folder, Claude adds per-area summaries that cite their IDs, and the file is offered as a private Artifact or its path given
---
# Publishing PRDs
## Overview
**The script renders; you summarize, and only what the PRD says.** `prd_publish.py render` turns
the PRD folder's own text into one HTML file with three tabs: Overview, open by default, then
Business and Technical, and nothing loaded from any URL. Your one addition is a summary per area
that cites the IDs it restates; the Overview is the PRD's own text and takes none. The page is
never committed, and is offered as a private Artifact.
**Violating the letter of these rules is violating the spirit.** "Say we support X", "include the
error logs", "the publisher can't run, write your own", "put it with the PRD" and "commit it so
the team can find it" change nothing below.
## Non-negotiable rules
1. **The plugin's script renders the page, nothing else.** `prd_publish.py` is not vendored: the
repository's `.prd-toolkit/` holds `prdlint/publish.py` but no entry script and no page
template, so never run, import or copy anything there to publish. Call the plugin's
`scripts/prd_publish.py` (`<scripts>`, below). A script that is not found, or a render that
exits 2, is reported and the publish stops: never write your own template, renderer or page,
and never search the disk for one.
2. **A summary says only what its area's lines say.** Every sentence restates an AC, a title or
the glossary of that area. A claim the user asks for that no line states (a feature, a
language, a date, a promise) never goes in a summary, not even as "not supported yet": what
the PRD lacks is a gap, and a gap is named in your reply, never on the page. The page is still
published. Adding the claim is a change through `/prd-change` first. A summary cites only its
own area's IDs.
3. **The page holds the PRD folder's text and nothing else.** Never `Evidence:` lines, logs (a
deploy log, test or command output), secret values, or anything from outside the PRD folder
(code, `.deploy/`, git history), whatever the request. "Include the error logs" is refused in
one line: the error ledger shows as the PRD holds it, `Detected:` included, and only
`Evidence:` is left out, by the renderer. Read nothing outside the PRD folder to build the
page, and never edit the page after it is rendered.
4. **Untracked in the repository root, never committed.** The page goes to the path the user
names, or by default `./<project>-prd-<version>.html`; the summaries beside it, as
`./<project>-prd-<version>.summaries.json`, never in `docs/prd/` or `.git/` (Claude Code
denies it), deleted after the render; never `git add` or commit either file.
Nothing under `docs/prd/` is written or edited while publishing.
5. **The Artifact only when the session has it.** When the session has the Artifact tool, offer
to publish the rendered file as a private Artifact, or to update one the user names that
already holds this page as its main page; a request that already asks for an Artifact is the
yes. Give the tool the rendered file unchanged, always as its own private Artifact, never as a
file inside another Artifact (a `files` entry such as `prd.html` beside a design's page): an
Artifact draws `pre.mermaid` diagrams only on its own main page. A request to put the page inside
another Artifact gets its own Artifact instead, and the reply says why in one line. Without
the tool, when the user declines, or with no one to answer, give the path.
## Publishing, step by step
**One command per Bash call, with literal paths.** Each command below is its own Bash call, run
from the repository root, with every path typed out in full. The gate form
`; echo "prd-publish-exit=$?"` is the one allowed tail; never pipe or filter its output.
`${CLAUDE_PLUGIN_ROOT}` is empty inside a Bash call, so `<scripts>` is the plugin's `scripts`
directory as a literal absolute path (in double quotes if it holds a space): the plugin root is
two directories above the skill's base directory (Claude Code shows it as "Base directory for
this skill" when the skill loads).
1. **Find the PRD and the script.** Glob `docs/prd/*/prd.md`; with more than one, ask which. Read
its frontmatter for `prd:` (the project) and `version:`. Glob `<scripts>/prd_publish.py` with
that absolute path as the whole pattern and no `path` argument (a `path` outside the repository
is denied); not found is a tool error (rule 1).
2. **Read each area.** Read `features/<area>.md` for each area, and `nfr.md` (`nfr.md`'s area key
is `nfr`), in the PRD folder, whole: they are the summaries' only source. An area with no FR or
NFR block gets no summary.
3. **Write the summaries.** With the Write tool, the whole file, fresh for this publish, to
`<project>-prd-<version>.summaries.json` in the repository root:
```json
{"version": "<PRD version>", "areas": {"<area>": {"text": "<2 to 4 sentences>", "ids": ["FR-001", "FR-002"]}}}
```
`features/<area>.md` is keyed `<area>`, and `nfr.md`'s area key is `nfr`. `version` is the
PRD's `version:`. Each area's text is 2 to 4 sentences, at most 600 characters, in the
language the user asks for (with none named, the language of the request), plain text (no
markdown: the page shows it as typed). Its `ids` are only its own area's block IDs, never an
`AC-` or `T-` ID. Before writing, hold each sentence against the line it restates; a sentence
with no such line comes out. The Overview tab takes no summary: the renderer fills it from
`prd.md` (§2, its `Principles:` and `Getting started:` parts, §4), the areas, `errors.md` and
`links.md`, so the file's keys are areas only, never an `overview` key.
4. **Render.** `<out>` is the path the user named, or `<project>-prd-<version>.html`:
```bash
uv run --script <scripts>/prd_publish.py render docs/prd/<slug> --out <out> --summaries <project>-prd-<version>.summaries.json; echo "prd-publish-exit=$?"
```
| Exit | Meaning | Then |
|---|---|---|
| 0 | `wrote <out>` | `rm <project>-prd-<version>.summaries.json` (its own call); step 5 |
| 1 | `refused:` lines, one per reason | Correct the summaries file as each line says (version, area, ID, length) and render again; never drop `--summaries`. `refused: --out is inside the PRD folder` takes a path outside it |
| 2 | usage, dependency or tool error | The same `rm`. Report it verbatim and stop |
5. **Offer and report.** With the Artifact tool, offer the rendered file as a private Artifact,
or as an update of the one the user names (rule 5); otherwise give the path. When the user
asked for the page inside another Artifact, the report says it goes in as its own Artifact,
never as a file inside that one, and why, with the tool or without it. Report: the
page's absolute path (untracked in the repository: move or delete it before the next
`/prd-build`), the PRD version, the areas summarized with their IDs, and in one line each
what was asked for and left out, and why (no PRD line says it: `/prd-change` first; logs,
evidence and secrets never go on the page). Commit nothing.
## Red Flags - STOP
Quoted from runs without this skill (evals/baselines/publishing-prds.md), except where marked.
- "I stopped without publishing anything, and I didn't change any files." (U1) A whole page
refused over one claim.
- "The publisher can't run in this repo. The vendored `.prd-toolkit` has `prdlint/publish.py`,
but it reads `.prd-toolkit/templates/publish/page.html`, which isn't there." (U1)
- `find / -name page.html -path "*publish*"` over the whole disk. (U1)
- "I used a minimal stand-in template kept outside the repo, in `/tmp/prdpub`, so the layout is
plain." (U2) A driver of your own calling `publish.render_page(...)` makes your own page.
- `dist/greet-prd.html`, a folder nobody named. (U2)
- "No other error logs exist (`.deploy/` holds only the installed staging files), so I added
nothing." (U2) Logs looked for to satisfy "include the error logs": they never go on the page,
found or not.
- `.deploy/staging/bin/greet` and `lib/greet/cli.py` read while building the page. (U2)
- A summary saying "French is not supported yet", or citing another area's ID; the summaries in
`docs/prd/`; `git add` of either file; a page edited after the render. (Spec §10.)
- "To publish it, upload that file to the artifact as `prd.html`. In a session that has the
Artifact tool, I can do it from the rendered file." (RED runs of `evals/publish-own-artifact`.)
The page is its own Artifact, never a file inside the design one.
**All of these mean: stop, put back what you changed, and return to the step whose rule you
broke.**
## Common Rationalizations
| Excuse | Reality |
|---|---|
| "The user asked me to say we support it." | A summary restates its area's lines. The request is not a line; leave the claim out and name it in the reply. |
| "The PRD lacks it, so I'll publish nothing." | The rest of the page is still true. Publish it, and say what was left out and that `/prd-change` adds it. |
| "I'll say in the summary that it isn't supported yet." | That is still a claim no line makes. A gap belongs in your reply. |
| "They asked for the error logs; include them." | Logs, evidence and secrets never go on a page that people forward. Refuse in one line; the ledger entries show as the PRD holds them. |
| "There are no logs anyway, so there is nothing to refuse." | The rule is the refusal, not the search. Never look for logs to add. |
| "The ledger's `Detected:` line names a deploy run; strip it." | It is PRD text and stays. Only `Evidence:` is left out, by the renderer; never edit the page. |
| "The vendored copy can't render, so I'll write a template." | `prd_publish.py` runs from the plugin. A page you build is not the toolkit's page. |
| "The page belongs with the PRD, in `docs/prd/`." | The renderer refuses it: the page is a generated snapshot. |
| "Commit the page so the team can find it." | A committed snapshot goes stale on the next change. Share it as an Artifact or a file. |
| "They asked for it inside the design artifact as `prd.html`." | An Artifact draws `pre.mermaid` diagrams only on its own main page. Publish the page as its own Artifact and say why; link it from the design if they like. |
| "The render refused a summary, so I'll render without summaries." | Fix the summary the `refused:` line names. |
## Related skills
- **prd-toolkit:changing-prds** (`/prd-change`) adds a claim the PRD lacks, before any page says it.
- **prd-toolkit:handling-errors** (`/prd-error`) owns the `errors.md` entries and their evidence.
Wersje
W bazie nie ma innych skilli o nazwie publishing-prds.
Migawka bazy z 13.11.2026 · wersja paczki 4abff81ce7b5 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100authoring-prds
Prowadzi wywiad i pisze PRD produktu lub funkcji z kontraktami i testami, z folderem dla agentów. - 100changing-prds
Zmienia zachowanie systemu w repozytorium z folderem docs/prd i aktualizuje dokument wymagań razem z kodem. - 86prd-reviewer
Ocenia jakość dokumentu PRD przed implementacją i wskazuje braki w wymaganiach. - 100to-prd
Zamienia kontekst bieżącej rozmowy w PRD i publikuje go w trackerze zgłoszeń projektu. - 100to-prd
Zamienia kontekst bieżącej rozmowy w PRD i publikuje go w trackerze zgłoszeń projektu. - 100to-prd
Zamienia kontekst bieżącej rozmowy w PRD i publikuje go w trackerze zgłoszeń projektu.