feat: add ATLAS knowledge base integration

[chipoto69]

Summary

• Add atlas_query tool with 5 actions: search, concept, insights, actions, stats
• Register tool in clawdis-tools.ts
• Add ATLAS skill documentation for agent usage
• Add ATLAS Bridge Guide for Phase 2 preparation (Push Path)

Phase 1: Pull Path (This PR)

Enables Clawdis agent to query the user's ATLAS personal knowledge base via HTTP API.

New Files

File	Description
src/agents/atlas-tool.ts	ATLAS query tool implementation
skills/atlas/SKILL.md	Agent documentation for ATLAS usage
docs/ATLAS_BRIDGE_GUIDE.md	Guide for ATLAS team to prepare Phase 2

Tool Actions

Action	Description	Required Params
search	Semantic search across knowledge	query
concept	Get concept details	concept_name or query
insights	Get curated insights	query (optional)
actions	Get pending action items	none
stats	Get knowledge base overview	none

Phase 2: Push Path (Future)

The included docs/ATLAS_BRIDGE_GUIDE.md documents how the ATLAS team should prepare for webhook-based event notifications (new insights, notes, content ingested, agent finished).

Testing

1. Start ATLAS: atlas serve (port 8888)
2. Start Clawdis gateway
3. Ask agent: "What do I know about [topic]?"

Environment Variables

ATLAS_URL=http://localhost:8888  # Default, configurable

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[steipete]

🦞 Codex Review

Findings

• ⚠️ Medium: No request timeout/abort; a stalled ATLAS call can hang the agent tool indefinitely. Recommend AbortSignal.timeout/AbortController with a default (match gateway 10s). (src/agents/atlas-tool.ts:71)
• ℹ️ Low: ATLAS_URL not trimmed/normalized; trailing slash/whitespace yields //api/... or invalid URL. Normalize once (trim + strip trailing /).
• ℹ️ Low: limit not sanitized; NaN/negative/float flows into query/body. Clamp + floor like other gateway option parsing.

Open Questions

• Do ATLAS endpoints accept limit on /api/actions and /api/search/insights? Guide lists bare endpoints.

Summary
Adds atlas_query tool, registers it in createClawdisTools, plus new skill and bridge guide.

Review by Clawd 🦞 via Codex

[steipete]

Thanks Rudlord! Not quite sure about this yet, need to spend more time researching atlas. Something local like the QMD system we already have would be nicer, or sth that can extend clawdis via hooks so this is user-addable.

pi has a great plugin system, might steal from that.

[steipete]

Wait I can't even sign up for this service? only book a call?

[steipete]

Thank you! This is better served as skill + cli for https://clawdhub.com/

[gabrielbernal]

Valid concern in theory. In practice it's a non-issue:

1. Fallback is valid and proven — openai/gpt-5.2 is configured with a working API key and has 21/21 success rate in production today.

2. Same end result with invalid fallbacks — with or without the fix, an invalid fallback model fails identically:

   • OLD: rate_limit → rotate 3 profiles (all rate limited, ~25s) → FailoverError → model-fallback → invalid fallback → fail
   • NEW: rate_limit → FailoverError → model-fallback → invalid fallback → fail

3. Profile rotation was not helping — production logs confirm all 3-4 rotations returned rate_limit before fallback triggered. Rate limits are per-org on Anthropic, not per-API-key.

The only scenario where rotation would help is if a profile maps to a different org with separate rate limits. This is a niche config better served by the model-fallback chain (which already tries different providers).