| name | ppt-research-trigger |
| description | Interact with the research routine itself — fire `deep` or `reset` runs from a Claude Code session without crafting curl by hand. |
PPT Research Trigger
routine-prompt.md § Special trigger payloads documents three modes:
text payload | Effect |
|---|
"" (empty) | Normal run — incremental since last_pr_seen / last_commit_sha / last_issue_seen. |
"deep" | 30-day catch-up sweep. Cursors only advance after all writes succeed — opportunistic catch-up, not a cursor reset. |
"reset" | Wipe last_pr_seen, last_commit_sha, last_issue_seen, seen_signals, and hotspot_history. Next run does an initial 14-day sweep. |
Today users open the README, find the routine's API trigger URL, and craft curl by hand. This skill makes the payloads discoverable and the curl template copy-pasteable.
When to invoke
You (or the user) want to fire a non-default routine run. Examples:
- Routine missed a few days while you were on vacation →
text: "deep"
state.json looks corrupted or you want to re-sweep the last two weeks → text: "reset"- Routine just shipped a hardening change and you want to confirm it converges on real recent activity →
text: "deep" once, then watch the brief.
Skip this skill for a normal daily run — that's scheduled at 05:00 local in claude.ai/code/routines and needs no manual prod.
What it gives you
- The exact URL + payload shapes (
{"text": "..."} JSON body)
- The credentials each trigger needs and where to fetch them
- The expected response shape so you know "fired" vs "failed"
- Cross-reference to where the routine acts on each payload
Inputs
- The routine's API trigger URL — copied from the routine's settings page at
https://claude.ai/code/routines → ppt-research → API trigger (only visible if API triggers are enabled on this routine).
- An API token scoped to fire that routine. Fetched once from the same settings page; treat as a secret.
Steps
- Fetch the trigger URL + token (one-time per routine):
- Open
https://claude.ai/code/routines
- Click
ppt-research → scroll to API trigger section
- Copy the URL (looks like
https://api.anthropic.com/v1/code/routines/<routine-id>/triggers)
- Generate a new token if one isn't already present; record it in your local secrets store (
pass, op, whichever you use). Don't commit it.
- Pick the payload based on what you want:
{"text": ""} → identical to the scheduled run{"text": "deep"} → 30-day backlog catch-up{"text": "reset"} → cursor wipe + next run is a 14-day initial sweep
- Fire the trigger — single
curl, plain JSON body:
curl -fsS \
-X POST "$ROUTINE_TRIGGER_URL" \
-H "Authorization: Bearer $CLAUDE_ROUTINE_TOKEN" \
-H "Content-Type: application/json" \
-d '{"text": "deep"}'
Response is a small JSON blob with the new run's id; the routine itself starts in the background. Watch https://claude.ai/code/routines/ppt-research/runs for progress.
- Verify the run committed — after ~2–5 min, check the latest commit on
main:
gh api repos/martin-janci/property-management/commits/main --jq '.commit.message' | head -1
For text: "reset", today's brief includes the literal line "State reset; cursors cleared." For text: "deep", the brief notes the 30-day window in Since last run.
Deterministic verification
test -n "$ROUTINE_TRIGGER_URL" && echo OK
test -n "$CLAUDE_ROUTINE_TOKEN" && echo OK
echo '{"text": "deep"}' | jq -e '.text' >/dev/null && echo OK
Smoke check (single command)
test -f .research/routine-prompt.md && grep -q 'Special trigger payloads' .research/routine-prompt.md
After-task verification
gh api repos/martin-janci/property-management/commits/main --jq '.commit.message' | head -1
Cross-references
