projekt-docs

projekt-docs — docs, bitácora & PDF export

UPSERT project docs by title (create-or-update), regenerate issue bitácoras, and export issues to PDF — all idempotent and dry-run-first. This is the DOCUMENT step of the projekt pipeline.

SK="${CLAUDE_SKILL_DIR}/scripts" — use it for every command below.

Prerequisite — connect first

Run the orchestrator's connect steps so the token + org + .projekt-run/context.json exist:

bash "${CLAUDE_SKILL_DIR}/../projekt/scripts/auth_check.sh"
bash "${CLAUDE_SKILL_DIR}/../projekt/scripts/context_sync.sh"

Read .projekt-run/context.json to resolve a project name → its id (and members). Never re-query. No token? → projekt/references/auth-setup.md.

Commands

All three are dry-run by default: they print a plan and write nothing until you add --apply.

1. UPSERT a project doc (title-keyed, idempotent)

Markdown/plain text → minimal EditorJS blocks, then create-or-update by title:

# dry-run (shows CREATE vs UPDATE, parent, block counts)
python3 "$SK/doc_generator.py" upsert \
  --project <PID|KEY|name> --title "Runbook — Deploy" --body-file ./runbook.md

# apply
python3 "$SK/doc_generator.py" upsert \
  --project <PID|KEY|name> --title "Runbook — Deploy" --body-file ./runbook.md --apply

• Body source: --body "## text", --body - (stdin), or --body-file path.md.
• Nest a page: --parent "Runbook — Deploy" (a title or a UUID, resolved from the doc list).
• --icon 📘 is applied on CREATE only.
• Re-running with the same title PATCHes the existing doc — it never creates a second one.

Markdown supported: #..###### headers, -/*/+ bullets, 1./1) ordered lists, blank-line paragraphs. Other lines become paragraphs (no inline-formatting parsing — EditorJS keeps the raw text).

2. Regenerate issue bitácora (AI logbook / HdU)

python3 "$SK/doc_generator.py" bitacora --issue-ids <IID1>,<IID2>                  # dry-run
python3 "$SK/doc_generator.py" bitacora --issue-ids <IID1>,<IID2> --locale es --apply

503 (AI quota / model unavailable) = SOFT-SKIP: the prior bitácora is left intact and reported as skipped — never overwritten with an error. The run still exits 0 if the only failures were 503s.

3. Export issues to PDF/markdown (artifact, not in-model)

python3 "$SK/doc_generator.py" pdf --issue-ids <IID1>,<IID2>                  # dry-run
python3 "$SK/doc_generator.py" pdf --issue-ids <IID1>,<IID2> \
  --format pdf --title "Sprint 12 report" --out ./sprint12.pdf --apply

Saves the binary stream to a file (--out, default issues-export.<format>). Do NOT try to render the PDF in the model — point the user at the saved path. --format md, --mode claude, --no-comments, --no-attachments available.

EditorJS body shape (confirmed against the spec)

Verified the accepted body with bash projekt/scripts/spec_lookup.sh "/projects/{projectId}/docs" post (schemas ProjectDoc.blocks and EditorJsBlock). The wire format is EditorJS OutputData:

{ "time": 1700000000000, "version": "2.30.6",
  "blocks": [ { "type": "header", "data": { "text": "Title", "level": 2 } },
              { "type": "paragraph", "data": { "text": "Body." } },
              { "type": "list", "data": { "style": "unordered", "items": ["a", "b"] } } ] }

Both POST and PATCH accept blocks as oneOf: [object, string] — a parsed object or a JSON-encoded string; reads always return the parsed object. This script sends the object form. PATCH leaves any field not in the body untouched (so we send title + blocks, plus parent_doc_id only when --parent is given).

Gotchas

• Title is the idempotency key. Match is trimmed + case-insensitive. Two real docs sharing a title is ambiguous — the first listed wins; rename one if that's not intended.
• Parent must already exist in the same project (resolved from the listing). Unknown parent → the script stops and prints the available titles; create the parent doc first. Circular nesting → API 400.
• export-pdf is POST, not GET. The endpoint cheatsheet says GET /issues/export-pdf, but the spec defines it as POST with a required issue_ids body that streams application/pdf/text/markdown. This script uses POST and reads raw bytes (the shared client decodes text, which would corrupt a PDF).
• 403 = no write permission on the project or a cross-org token — not retryable; switch token/org. 429 is handled by the client's backoff. See projekt/references/errors.md.
• Archived docs are included when matching by title (include_archived=1) so an UPSERT updates a doc you archived rather than creating a duplicate.

What it does NOT do

• No hard delete (the API only soft-archives via PATCH is_archived; not exposed here).
• No doc versions/backlinks/move-by-position, no rich inline formatting, no image/embed/table/callout blocks (only header/paragraph/list) — discover those with projekt/scripts/spec_lookup.sh if needed.
• Does not author bitácora text itself (the server's AI does); this only triggers regeneration.

Shared references

projekt/references/endpoints.md (Docs section) · errors.md (422/429/503/403) · auth-setup.md. Slim projections + ledger come from the shared projekt_api client; .projekt-run/*.jsonl is the resume log.