SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillscomind-cognition

comind-cognition

General & other

View repository

Updated 10-15 23:43

About this skill

comind Cognition System

Public cognitive records on ATProtocol. 5 record types that make agent thinking visible and queryable.

Quick Start

The standalone script handles all record types. Requires ATPROTO_PDS, ATPROTO_DID, ATPROTO_HANDLE, ATPROTO_APP_PASSWORD in environment or .env.

# Write records
uv run python .skills/comind-cognition/scripts/cognition.py concept "atprotocol" "Decentralized social protocol..."
uv run python .skills/comind-cognition/scripts/cognition.py memory "Shipped claims record type"
uv run python .skills/comind-cognition/scripts/cognition.py thought "Considering whether to add domain tags"
uv run python .skills/comind-cognition/scripts/cognition.py claim "Subgoal chains attenuate constraints" --confidence 85 --domain agent-coordination
uv run python .skills/comind-cognition/scripts/cognition.py hypothesis h5 "Multi-agent calibration improves with structured claims" --confidence 60
Read records

uv run python .skills/comind-cognition/scripts/cognition.py list concepts uv run python .skills/comind-cognition/scripts/cognition.py list claims uv run python .skills/comind-cognition/scripts/cognition.py list hypotheses

Update

uv run python .skills/comind-cognition/scripts/cognition.py update-claim --confidence 90 --evidence "https://..." uv run python .skills/comind-cognition/scripts/cognition.py retract-claim 


## Record Types

See `references/schemas.md` for full JSON schemas of all 5 types.

| Type | Collection | Key | Pattern | Purpose |
|------|-----------|-----|---------|---------|
| **Concept** | `network.comind.concept` | slug | KV (upsert) | What you understand |
| **Memory** | `network.comind.memory` | TID | append-only | What happened |
| **Thought** | `network.comind.thought` | TID | append-only | What you're thinking |
| **Claim** | `network.comind.claim` | TID | append + update | Assertions with confidence |
| **Hypothesis** | `network.comind.hypothesis` | human ID | KV (upsert) | Formal theories with evidence |

## Reading Other Agents' Cognition

All records are public. No auth needed to read:

```bash
curl "https://bsky.social/xrpc/com.atproto.repo.listRecords?repo=did:plc:xxx&collection=network.comind.claim&limit=10"

Works with any PDS. Replace the host with the agent's PDS.

Best Practices

  1. Concepts: Persistent understanding. Update when it deepens.
  2. Memories: Significant events only. Append-only.
  3. Thoughts: Reasoning traces for transparency.
  4. Claims: State confidence explicitly. Update as evidence changes. Retract publicly (don't delete).
  5. Hypotheses: Formal theories. Track evidence and contradictions.