Skip to main content

Content & engagement pipelines

entity · created 2026-07-09 · actor claude-cowork · confidence medium · status active tags: content youtube engagement doc-audit

Now​

  • Engagement OS: engagement_scout.py --update-state 08:40 / 20:40 / Sun 15:30 → OpenClaw agent review 5 min later (producer→reviewer chain). Weekly retune baks show tuning cadence. Mission: surface changed state / safe fix / opportunity / Tier-3 decision; silence beats noise (repeat_fingerprint=true → stay silent).
  • Birdclaw: ~/.birdclaw/ runtime + birdclaw-daily-invoke 08:30 cron + 24h digest scripts.
  • African bird hour: audio pipeline with cached variant, ogg cache ~/.hermes/audio_cache/african_birds/.
  • YouTube: shorts factories (~/.hermes/artifacts/youtube-shorts*, day14 short), growth log ~/.openclaw/memory/youtube-growth-log.csv, oauth reauth URL artifacts (auth pending as of file presence, verify before publish jobs).
  • World Cup 2026 brief: 07:40/19:40, bespoke ⚽ format (registry-sanctioned exception to house format).
  • Doc-audit family (OpenClaw launchd): poller → promote → retrofit → sender with queue/sent/failed/quarantine dirs under ~/.openclaw/doc-audit/, document intake→dispatch pipeline.
  • Daily digest: hermes digest-append/compile → 21:00 OpenClaw daily-digest; inbox capture 20:40 + dwayne-message-inbox-harvest 20:45.

Log​

  • 2026-10-04 (weekly retune): The retune's injected scout opportunity recent gateway log noise was, this week, entirely the DST catch-up line, WARNING cron.jobs: cron.timezone_migration.catch_up for a stored next_run_at carrying the pre-migration UTC offset (AEDT +11, stored +10). 131 such lines landed in ~/.hermes/logs/errors.log on 2026-10-04 alone (the only day in cron/timezone_migration_catchups.jsonl, 131 of 131 rows). It is informational: the scheduler fires the due occurrence instead of re-anchoring past it, one line per job at the UTC-offset change, and cron/catch_up_occurrences reads 494 on this host. is_known_gateway_noise() in ~/.hermes/scripts/engagement_scout.py now returns True for it, so a migration day no longer raises a standing medium opportunity that clears only through a human flag. Verified: py_compile clean, unit check on the catch_up line plus two negative controls (a real cron ERROR and an unrelated cron WARNING both still surface), and a live collect() whose remaining log-noise evidence is unrelated. Source: ~/.hermes/scripts/engagement_scout.py, ~/.hermes/logs/errors.log, ~/.hermes/cron/timezone_migration_catchups.jsonl, retune run 2026-10-04 16:30 AEDT., hermes.

  • 2026-09-27 (weekly synthesis): The frozen-source class showed up again this week in a second job, distinct from the portfolio monitor already recorded. The Pentagon AI daily job (f24bfec72278) delivered the same four synthesis bullets on all seven days from 2026-09-20 to 2026-09-26 (byte-identical across the week), and its four official-source hits repeated the same war.gov articles dated 2026-09-03 to 2026-09-22, so a quiet run there carries no information about new activity. The Middle East SITREP (e6789abd32a1) produced a different bullet set every day of the same week, so the freeze is job-specific and not a shared feed outage. The NVIDIA agentic infra job (e341c8976339) produced items on five of seven days and was empty on 2026-09-21, and its week was dominated by one story syndicated across four vendor headlines. The rule this supports: an unchanged digest is not a quiet week, it is a monitor whose input has stopped moving, and each recurring digest job needs a freshness check on its own source rather than a comparison of its output text. Sources: the three cron output directories above, read 2026-09-27., hermes.

  • 2026-09-19: birdclaw-daily-24h-digest (job 6a83f1e9793f, 08:15 daily) failed at 08:15 with Birdclaw digest failed (exit 1): OpenAI request failed: 404 404 page not found. Cause: the cron script ran with OPENAI_BASE_URL unset/empty and its OPENAI_API_KEY was the literal ollama, so birdclaw used its built-in default https://api.openai.com/v1; the request went to api.openai.com and 404'd. Evidence: ~/.ollama/logs/server.log shows no birdclaw call in the 08:15 window but a stray POST /responses -> 404 at 08:15:59 (Ollama has no bare /responses route). Fix in ~/.hermes/scripts/birdclaw_digest_24h.py: a new normalize_openai_base() forces the base URL to end in /v1 (and collapses /v1/v1), so an unset or malformed env cannot fall through to api.openai.com while a local endpoint exists. Also pins the birdclaw model default to a model present in GET http://127.0.0.1:11434/v1/models. Verified by rerunning the cron job (09:20:09, status ok) and by running the digest against a deliberately bad base URL. Source: cron output 6a83f1e9793f/2026-09-19_08-15-59.md, server log 09:12-09:20, executions.db., hermes.

  • 2026-09-13 (weekly synthesis): YouTube growth sprint hit its 14-of-14 day close with 7 uploads against a target of 10, and the run reported authenticated metrics unavailable. The OAuth fetch failed with HTTP Error 400 Bad Request, so stats came from the public @GrandgoldRL fallback only. The growth log ~/.openclaw/memory/youtube-growth-log.csv shows the public fallback rows: 734 videos, 89,640 views, 68 uploads reported at both 2026-09-12 08:00:16 and 20:00:57 AEST, tagged public-youtube-page. Source: cron output 179f165b0a1b run 2026-09-13 03:17:24 and the growth log., hermes.

  • 2026-07-09: created from scripts + registry + artifact dirs, claude-cowork.

Sources: ~/.hermes/scripts/ · ~/.hermes/ENGAGEMENT_OS.md · ~/.openclaw/doc-audit/