process

Files (1)

• SKILL.md

SKILL.md (first lines)

Analysis contract (rules + output schema) for this run:

!`cd "${CLAUDE_PLUGIN_ROOT}" && npx tsx --env-file-if-exists=.env scripts/print-analyze-contract.ts`

Process all new recordings. Publish for real only when `$ARGUMENTS` contains `--commit`; otherwise
do a dry run (analyze and report, but do not publish or mark).

1. List recordings to a file (pass through `--days N` if present):
   ```bash
   cd "${CLAUDE_PLUGIN_ROOT}" && npx tsx --env-file-if-exists=.env scripts/list.ts --json $ARGUMENTS
   ```
2. Read `artifacts/recordings.json` and the ledger `state/ledger.json` (treat a missing ledger as
   an empty list). The NEW recordings are those that are `complete` and whose `id` is not in the
   ledger, regardless of who recorded them - the listing includes meetings recorded by teammates,
   with `recorderName` identifying the recorder.
3. For each NEW recording id, in order, isolating failures (a bad recording must not stop the rest):
   a. `npx tsx --env-file-if-exists=.env scripts/download.ts <id>`
   b. Read `artifacts/<id>.transcript.json` + `artifacts/<id>.meta.json`, produce the action-plan
      JSON per the contract above (stamp recordingId/recordingName/recordingDate/recorderName from
      meta; recorderName is null when meta has none), write `artifacts/<id>.plan.json`.
   c. `npx tsx --env-file-if-exists=.env scripts/validate-plan.ts <id>` (fix and re-write on failure).
   d. If `--commit`:
      - Notion: read the cached destination
        `npx tsx --env-file-if-exists=.env scripts/config-get.ts notion.databaseId`.
        If it is EMPTY, skip publishing this recording and do NOT mark it - report that no Notion
        destination is cached (the user must run `/loom-daddy:publish` once interactively to set
        it). If present, create the page via the available Notion MCP create-pages tool, using the
        same layout as the publish skill (title, a `Recorded by: <recorderName>` paragraph first
        when the plan has a non-null `recorderName`, Summary, Decisions with `?t=` links, Action
        Items table, Open Questions).
      - Record the confirmed Notion publish using the page URL the create-pages tool returned:
        `npx tsx --env-file-if-exists=.env scripts/record-publish.ts <id> --url <pageUrl>`.
        If the create-pages call did not return a page URL, the publish did not happen - do NOT
        record it and do NOT mark; leave the recording for the next run.
      - Todoist: read `npx tsx --env-file-if-exists=.env scripts/config-get.ts todoist.projectId`.
        If EMPTY, skip task creation (report once for the run; the user configures it via
        `/loom-daddy:publish` interactively). Otherwise run
        `npx tsx --env-file-if-exists=.env scripts/todoist-payload.ts <id>`; if the JSON has
        `"skip": true`, continue. Else create the parent task then the subtasks via the Todoist
        MCP add-tasks tool (subtasks use `parentId` = the parent task id), and record with
        `npx tsx --env-file-if-exists=.env scripts/record-todoist.ts <id> --task-id <parentId> --count <n>`.
        If the Todoist MCP is unavailable or returns no task id, do NOT record and continue;
        list the recording as "tasks unsent" in the summary. Todoist never blocks Drive or
        marking.
      - Drive: `npx tsx --env-file-if-exists=.env scripts/publish.ts <id> --drive-only --commit`
        (skips itself when Drive is not configured).
      - Then `npx tsx --env-file-if-exists=.env scripts/mark.ts <id>`. mark.ts refuses to mark
        unless the Notion publish receipt from the previous step exists, so a missing or failed
        Notion publish leaves the recording unmarked for retry.
4. Report a per-recording summary: processed, skipped (already in ledger or no Notion destination),
   failed ids, and any recordings whose Todoist tasks are unsent.

Note: this skill is model-invocable so the local scheduler can trigger it headlessly via
`claude -p`. A headless run reaches the claude.ai Notion MCP (verified), so publishing uses the
MCP in both the interactive and scheduled paths. Do not advance any time watermark beyond what the
scripts already persist.

From fontezbrooks/loomDaddy, MIT.

More from fontezbrooks/loomDaddy