| name | agent-learning-ctx-evidence |
| description | Use ctx-backed local agent-history search as bounded evidence references while creating, validating, or consolidating voku/agent-learning findings. |
Agent Learning ctx Evidence
Use this skill when creating, validating, or consolidating findings and prior agent sessions may contain relevant decisions, failed attempts, commands, review context, or validation history.
This skill depends conceptually on ctx-agent-history-search. Use that ctx skill for the detailed search workflow; use this skill to preserve the agent-learning boundary.
Workflow
- Check whether ctx is available and indexed:
ctx status
ctx sources
- Search for focused historical context:
ctx search "<task / failure / module / command>"
- Inspect before relying on a result:
ctx show event <ctx-event-id> --window 5
ctx locate event <ctx-event-id>
- Create or update an agent-learning finding with
agent_history_reference evidence.
Evidence Rules
- Set
type to agent_history_reference.
- Set
source to ctx.
- Record
query, retrieved_at, summary, and at least one of ctx_session_id or ctx_event_id.
- Use
verification_status=found for search hits that were not opened.
- Use
verification_status=inspected only after reviewing the event or session.
- Use
verification_status=rejected for irrelevant or wrong hits.
- Use
verification_status=stale for old context that may no longer apply.
Do not paste raw transcript text unless it was explicitly reviewed and is necessary. Do not store large JSON payloads, secrets, tokens, or private-path-heavy excerpts. Do not mark ctx-derived findings as validated without independent project evidence.
ctx helps find what happened before; agent-learning decides what deserves to survive.