chip-subconscious

name	chip-subconscious
description	Read-only sensing layer for Hermes/OpenClaw-style agent workspaces. Use when you want scheduled discovery walks, signal scoring, pending intents, Telegram topic escalation, and human approval buttons before any build work.
created_by	public-template
license	MIT

chip-subconscious

A public, secret-free SUBCONSCIOUS skill for Hermes/OpenClaw-style workspaces.

It watches read-only surfaces, scores repeated signals, creates pending intents, and escalates them to chat topics for human approval. It does not build, deploy, restart services, edit configs, rotate keys, or approve itself.

Trigger this skill when

• You want a separate read-only sensing layer for Hermes/OpenClaw/memory infrastructure.
• You want scheduled “walks” that notice repeated pains, blockers, and reusable opportunities.
• You want pending intents with ✅ Yes / ❌ No buttons before implementation starts.
• You need Telegram forum topics or equivalent chat branches for escalation lanes.

State machine

observed -> scored -> pending_intent -> pending_approval -> approved -> planned -> building -> qa -> done/rejected/archived

SUBCONSCIOUS stops at pending_approval.

Required room lanes

• signal_board — durable board of current signals.
• walk_logs — scheduled read-only walk summaries.
• pending_intents — cards awaiting human approval.
• approved_builds — approved intent files and build packets.
• archive — rejected/cooled/old items.

See references/telegram-escalation.md.

Quick start

git clone https://github.com/evgyur/chip-subconscious.git
cd chip-subconscious
python3 scripts/subc_init_room.py \
  --chat-id REPLACE_WITH_TELEGRAM_CHAT_ID \
  --topic-ids 1,2,3,4,5
python3 scripts/subc_walk.py --source all --no-publish
python3 scripts/subc_score.py
python3 scripts/subc_intents.py --max 3
python3 scripts/subc_validate.py

Optional Telegram publish requires TELEGRAM_BOT_TOKEN in the environment or ~/.hermes/.env and a configured telegram_topics.json.

Source adapters

• HermesAdapter checks Hermes config, skills, sessions, and the SUBCONSCIOUS room.
• OpenClawAdapter checks configured OpenClaw/GoClaw service names and skill paths with read-only systemctl is-active.
• Mem0gAdapter checks a localhost health URL, configured services, and configured evidence paths.
• ProjectFlowAdapter finds STATE.yaml files under the workspace root.

All adapter settings are environment-configurable. See references/openclaw-hermes-sources.md.

Guardrails

Forbidden for SUBCONSCIOUS:

• service restart/reload/start/stop
• deploy
• edit config
• key/grant/actor mutation
• writing production data
• reading/pasting raw secrets
• approval/self-approval
• build start

Allowed:

• read-only discovery
• redacted logs
• scoring
• pending intent creation
• escalation to chat topics

Commands

# Initialize room
python3 scripts/subc_init_room.py --chat-id CHAT_ID --topic-ids BOARD,WALKS,INTENTS,BUILDS,ARCHIVE

# Run read-only discovery
python3 scripts/subc_walk.py --source all --no-publish --limit 25

# Score observations and create intents
python3 scripts/subc_score.py
python3 scripts/subc_intents.py --max 3

# Publish pending cards to Telegram topic
python3 scripts/subc_publish_pending.py

# Validate room and run tests
python3 scripts/subc_validate.py
python3 -m unittest discover -s tests -v

References

• references/openclaw-hermes-sources.md — how source adapters collect info.
• references/telegram-escalation.md — chat branches, buttons, escalation flow.
• references/security.md — public-safe setup and redaction rules.
• references/operations.md — scheduled runbook.