create-canon

Create a canon

Canonify's authoring gate. A canon is one canonical-pattern doc (a .md); the root manifest CANONIFY.md indexes every canon by a one-line summary, and the lifecycle gates (plan / build / commit) route off those summaries. Where they consume the canon, this gate grows it: point it at a file, service, or design element and it writes a new canon and wires it into CANONIFY.md so the other gates can immediately route to it.

It codifies the workflow for authoring a canon: read the real code, cite file:line, draft from evidence, and ask only for what the code can't tell you.

How to run

1. Identify the target. The user points at one of:

   • a file / class (a path or a type name),
   • a service / integration (e.g. a third-party API client, a *Service wrapper),
   • a design element (a UI pattern + its CSS/JS source). If nothing is named, ask once: "what file, service, or element should I turn into a canon?" Don't guess.

2. Place it. Decide the category + destination from CANONIFY.md's blocks and the existing docs/ folders - the repo defines its own categories. Don't impose a fixed taxonomy; read what categories already exist (the blocks in CANONIFY.md, the subfolders under docs/) and slot the canon into the matching one, creating a new category only when none fits. The destination is docs/<category>/<topic>.md. Infer the obvious case; confirm the destination path in one line.

3. Index the target (the evidence pass). Read the target in full and gather the proof every claim will cite:

   • the wrapper / class and its public surface,
   • how it's wired up / registered / constructed (DI, factory, entry point),
   • config keys / settings and where they're read (file:line),
   • real call sites (grep the repo),
   • related types / canons it overlaps with. For a design element also read its source partial(s), any vendor library it depends on, and a reference template / view that uses it.

4. Ask only the gaps. Draft from the code first, then ask a SHORT set of questions for what the code cannot reveal:

   • the one-line scope / title (confirm),
   • gotchas / footguns the author knows that aren't visible in code,
   • anything to deliberately defer to another canon (overlap),
   • (design) a production screen or tokens to match. Keep it to a few questions. Don't interrogate - prefer reasonable defaults and confirm.

5. Write the canon following the doc-style template that ships with Canonify (${CLAUDE_PLUGIN_ROOT}/templates/doc-style.md). The core rules, inlined so this gate is self-contained:

   • Cite every factual claim with file_path:line_number. If you can't point at it in the code, don't write it. Never invent a method name, config key, or behavior.
   • ASCII punctuation only (- not em-dash, straight quotes, -> not arrow, ... not ellipsis).
   • Terse and project-specific. Document THIS codebase's usage, not the library's general theory.
   • Section skeleton, in order: title + one-line intro; a living-standard + routing note; a reference-implementation table (the files to copy from); the pattern (a real excerpt with file:line); config / API surface; recipes (real call sites); gotchas; references.
   • verified: frontmatter marker. Open the canon with a YAML frontmatter line verified: <git-sha> recording the commit at which the canon was confirmed against the code. Set it here at authoring time; the Doctor gate reads it to flag drift. One SHA, set by the gates - never hand-maintained.
   • For a UI / design canon, add a Mockup recipe section: a static, copy-pasteable snippet that reproduces the element so a mockup matches production. If the element needs a JS library a static mockup can't run, give a representative static approximation.

6. Register the canon (required). An unregistered canon is invisible to routing, so do both:

   • add a one-line summary to the category's docs/<category>/INDEX.md,
   • add the same summary - as a ### docs/<category>/<topic>.md block - to CANONIFY.md under the right category. The summary is what the gates match on: name the key types / area so a future task or diff that touches it matches here.

7. Report. The new canon's path, what was indexed, the INDEX + CANONIFY.md edits, and any gaps the author still needs to fill (e.g. a gotcha you couldn't verify from code).

Don't do

• Don't skip registration. Writing the canon without updating CANONIFY.md + the INDEX leaves it un-routable - the most common and most damaging miss.
• Don't invent. No claim without a file:line. If you can't verify it, ask or omit it.
• Don't over-ask. Draft from the code; ask only what the code can't tell you.
• Don't paste vendor theory. Document THIS codebase's usage, not the library's general docs.
• Don't duplicate. If a canon already covers the area, update it instead of writing a second one.

Output template

## /canonify:create-canon - <target>

Indexed: <files read + greps run>
Destination: docs/<category>/<topic>.md  (category: <chosen from CANONIFY.md / existing docs/ folders>)

Questions (gaps only):
1. <question the code can't answer>
2. ...

[after answers -> write the canon, then:]

Created: docs/<category>/<topic>.md  (the new canon, <N> lines, every claim cited file:line)
Registered:
- docs/<category>/INDEX.md   (+1 summary line)
- CANONIFY.md                (+1 block under "<Category>")

Gaps to fill: <anything unverified the author should confirm>