agent-teams-command
Use when work has genuinely independent streams or distinct builder, evaluator, domain, and integration roles that require…
Use when a completed work session should yield durable concepts, corrections, decisions, reusable patterns, and a traceable next action.
$ npx -y skills add Mark393295827/third-brain-v7-skills --skill session-learn --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/session-learnContext preview
The summary Claude sees to decide when to auto-load this skill.
Use when a completed work session should yield durable concepts, corrections, decisions, reusable patterns, and a traceable next action.
name: session-learn description: Use when a completed work session should yield durable concepts, corrections, decisions, reusable patterns, and a traceable next action. metadata: version: "8.1.0" updated: "2026-08-18" profile: "stateful" assumes: "The session transcript, artifacts, or execution receipts are available." conflicts_with: "Claiming automatic capture without a configured hook, storing raw noise as knowledge, or changing immutable sources."
<skill_contract> <input>Completed session objective, actions, artifacts, verification receipts, errors, decisions, and destination paths.</input> <output>Deduplicated durable deltas for concepts, entities, corrections, patterns, ideas, decisions, and gaps.</output> <done>Each retained delta is formatted, linked, logged, provenance-preserving, and paired with a traceable next action when unresolved.</done> <non_goals>Automatic-capture claims without a hook, raw transcript storage, conversational filler, or immutable-source edits.</non_goals>
Close a session by extracting only reusable deltas. Invocation must be explicit or performed by a verified external hook; this skill never claims it ran automatically.
Provide: session objective, actions, outputs, verification receipts, errors, decisions, and destination paths. Optional: existing notes for deduplication.
<intake>
Establish the session boundary and compare intended versus observed result. Ignore conversational filler and separate execution evidence from retrospective interpretation.
</intake>
<unknowns_gate>
If the session outcome or evidence is unavailable, return `INSUFFICIENT_EVIDENCE`. Ask for a missing artifact only when it determines whether a lesson is valid; otherwise preserve it as an unresolved gap.
</unknowns_gate>
<execute>
Scan for seven signal types:
1. **Concept** — a stable mechanism worth linking. 2. **Entity** — a person, system, project, or tool requiring durable context. 3. **Correction** — a prior belief or procedure disproved by evidence. 4. **Pattern** — a reusable Trigger -> Execute -> Verify -> State sequence. 5. **Idea** — an untested possibility, explicitly provisional. 6. **Decision** — choice, rationale, alternatives, owner, and review condition. 7. **Gap** — an unknown with a probe or escalation path.
Apply Closure Protocol: **Format** the smallest durable note, **Link** it to source/project/concepts, and **Log** the write plus verification. Merge semantic duplicates; preserve source immutability and provenance.
</execute>
<evaluate>
For each candidate ask: Is it new, reusable, traceable, and decision-relevant? Reject session-specific trivia and unsupported generalizations. Confirm each write exists and links resolve before reporting closure.
</evaluate>
<state_contract>
Persist `{run_id, status, attempt, budget, evidence, unknowns, last_error, next_action}` plus session id, extracted candidates, accepted/rejected reasons, write paths, links, and closure receipt. Append corrections; do not erase superseded beliefs.
</state_contract>
Return `status`, `result` (accepted knowledge deltas), `evidence` (session receipts and write checks), `unknowns`, and `next_action`.
</skill_contract>
Verification-first Agent Skills for Codex OS, with optional compatibility adapters for Claude Code, Gemini, Cursor, and Windsurf.
Use when work has genuinely independent streams or distinct builder, evaluator, domain, and integration roles that require…
Use when designing or refactoring a model-native engineering workflow with bounded autonomy, probes, custom evaluation, durable…
Use when property-service operations need an AI plus ontology plus DMAIC design for work orders, dispatch, quotes, evidence, CTQ…
Use when a personal or team operating system needs a bounded redesign using Four-C, closed-loop controls, 70/30 allocation, 3B…
Use when a goal must be converted into a repeatable behavior, cue, SOP, review cadence, and identity-aligned reinforcement.
Use when source material must be transformed into a compact, evidence-aware model for learning, decisions, or an Obsidian concept…