0 Fork 0

Files

T

History

beantownbabeandClaude Sonnet 5 9edc52cd15 Change Conference Scout cadence from weekly to monthly

Cron changed from every Monday (0 6 * * 1) to the 1st of every month
(0 6 1 * *), next run 2026-09-01. Updated the tracked config.json,
its README, and the dashboard's own description text to match.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

2026-08-19 09:49:44 +02:00

config.json

Change Conference Scout cadence from weekly to monthly

2026-08-19 09:49:44 +02:00

README.md

Change Conference Scout cadence from weekly to monthly

2026-08-19 09:49:44 +02:00

README.md

Conference Scout

A scheduled cloud agent (Claude Code "routine") that searches the web on the 1st of every month for conferences, trade shows, seminars, and networking events relevant to TXproducts, and produces a report for manual review. It does not write to this repo, any calendar, or any file — a person reviews its report and decides what to add to the Opportunities tab's Events list themselves.

Where it actually runs

This agent does not run on TXproducts infrastructure and has no server-side component anywhere. It runs entirely as a scheduled routine on Anthropic's cloud (via the Claude Code "routines" feature), triggered by cron and executed in an isolated sandbox with WebSearch/WebFetch access only. There is no code to deploy or host — the routine's entire behavior is defined by the prompt in config.json.

Live routine: https://claude.ai/code/routines/trig_01XRb81y9pAWDube5kjRUvGT

What's in this folder

config.json — the routine's definition: schedule, model, allowed tools, and the full prompt that defines its scope (what counts as "relevant," what's explicitly excluded, geographic coverage, output format). This is the actual source of truth for what the agent does — read the prompt to understand its behavior.

This file is a manually-maintained mirror, not a live sync. If someone edits the routine directly on claude.ai, this file goes stale until someone updates it here to match. There is currently no automation keeping the two in sync.

How to update it

Change the routine itself first (via the Claude Code RemoteTrigger tool's update action, or the claude.ai UI), and confirm the change actually took by fetching it back (RemoteTrigger get action).
Copy the confirmed live prompt/schedule/config into config.json here, update _live_reference.last_synced_from_live, and commit.

How to recreate it if it's ever deleted

config.json's top-level shape (name, cron_expression, enabled, job_config) matches the RemoteTrigger create action's request body directly — it can be passed to recreate the routine from scratch (you'll get a new trigger_id; update _live_reference afterward).

Why results aren't fully automatic

Each run is a fresh, isolated cloud session with no memory of prior runs and no write access to this repo (by design — see the prompt's explicit "do NOT add anything to any file, calendar, or repository" instruction). Syncing a run's results into the Opportunities tab's Conference Scout view (opportunities/opportunities.js, SCOUT_RESULTS) is a manual step done by whoever reviews that week's report.

Known limitation (as of 2026-08-17)

The routine's sandbox has consistently blocked all WebFetch calls to external sites — even neutral test domains like wikipedia.org — via a network egress policy, despite the prompt instructing it to verify facts via WebFetch on official event pages. In practice it falls back to cross-referencing multiple independent WebSearch results instead. Every synced report should be read as "search-corroborated," not "primary-source-verified," until this is resolved (if it ever can be, since it appears to be an environment-level restriction rather than a prompt issue).