name: Structured Experience Memory description: Use this skill whenever tackling non-trivial coding tasks, debugging bugs, fixing test failures, refactoring components, or recording verified task outcomes. Provides structured validated experience retrieval (including DO NOT REPEAT failure warnings), candidate experience recording, evidence-backed VerificationGate promotion, environment compatibility checking, and cross-agent experience transfer. version: 1.0.0

CogniCore Structured Validated Experience Memory

CogniCore provides a verified, structured memory layer that persists across Claude Code sessions. It captures what problems occurred, what approaches failed (and why), what approach succeeded, and verifiable execution evidence (test commands and exit codes).

Architecture & Workflow

Claude Code Session
       │
       ▼
1. cognicore_recall_experience(query)
       │
       ├──► Inspect failure_warnings (DO NOT REPEAT)
       └──► Inspect experiences (VERIFIED solutions)
       │
       ▼
2. Execute Coding Task (try approaches, run tests)
       │
       ▼
3. cognicore_record_experience(task, problem, solution, attempts_json, why_it_worked)
       │  (Recorded as CANDIDATE)
       ▼
4. cognicore_verify_experience(experience_id, evidence_json)
       │
       ├──► Exit Code == 0 with evidence ──► Promoted to VERIFIED
       └──► Exit Code != 0 or no evidence ──► Remains CANDIDATE / REJECTED

The 5 MCP Tools

1. cognicore_recall_experience

Call this BEFORE attempting a non-trivial problem or fix.

Parameters:

• query: Description of the bug, feature, or error message (e.g. "Fix ZeroDivisionError in math_utils.py calculate_ratio").
• include_failures: true (default) to receive explicit DO NOT REPEAT warnings from previous failed attempts.
• require_verified: false to see all relevant experiences or true for only verified ones.
• python_version / framework / dependencies_json: Current runtime environment for compatibility matching.

How to act on recall results:

• If failure_warnings are present: DO NOT REPEAT those specific approaches.
• If experiences with status verified are present: Use the proven solution pattern as strong guidance.

2. cognicore_record_experience

Call this AFTER completing a coding task where you explored approaches or solved a non-trivial bug.

Parameters:

• task: High-level goal (e.g. "Fix division by zero bug").
• problem: Specific failure mode or error symptom.
• solution: Concise description of the working fix.
• why_it_worked: Root cause explanation.
• attempts_json: JSON array documenting all attempts made during the session:

``json [ {"approach": "Multiply denominator by 1.0", "outcome": "failure", "reason": "Still divides by 0 on zero input"}, {"approach": "Wrap caller in try/except", "outcome": "failure", "reason": "Violates function API contract"}, {"approach": "Add guard clause for zero denominator", "outcome": "success", "reason": "Safely handles boundary condition"} ] ``

• repository_id: Repository identifier.
• commit: Git commit hash.
• python_version / dependencies_json: Environment metadata.

3. cognicore_verify_experience

Call this to promote a candidate experience to VERIFIED.

[!CRITICAL]

Experiences are NEVER verified simply because of a text claim. Verification requires actual command execution evidence with exit_code == 0.

Parameters:

• experience_id: The ID returned by cognicore_record_experience.
• evidence_json: JSON array of execution evidence:

``json [ { "command": "pytest tests/test_math.py -q", "exit_code": 0, "stdout_hash": "c0ffee1234", "commit": "abc1234" } ] ``

4. cognicore_check_experience

Call this when working with past experiences in an upgraded environment or when dependencies have changed.

Parameters:

• experience_id: The experience ID.
• python_version / framework / dependencies_json: The new environment details.

Outcome: If dependencies or versions have breaking changes (e.g., lib v1.0.0 vs lib v2.0.0), CogniCore returns stale: true and status: "revalidation_required". Revalidate with fresh tests before relying on it as truth.

5. cognicore_share_experience

Call this to securely transfer verified experiences and failure warnings to another agent or workspace. Only verified experiences with intact provenance are transferred.

Best Practices

1. Never dump full conversation logs: Only record concise problem statements, structured attempts, root causes, and verification proof.
2. Never hide failures: Failed attempts are first-class memory records that prevent future sessions from wasting time on dead ends.
3. Always verify with test evidence: Always run the relevant test command and supply its exit code to the verification gate.