titan-memory-prune

# titan-memory-prune

## Intent
Use this skill to review Titan memory records for retention, masking, tombstone, or pruning proposals.

## Trigger boundary
Use this skill when:
- a Titan memory record may be stale, sensitive, or wrong
- a recall result needs redaction handling
- a bearer lineage record should preserve fall or correction without erasure

Do not use this skill when:
- the request asks for silent deletion
- the record lacks source or authority metadata
- pruning would hide an important lineage event

## Inputs
- memory record id
- source refs
- reason for prune request
- retention rule
- operator or owner confirmation state

## Outputs
- redaction proposal
- tombstone proposal
- records to keep unchanged
- reason and source refs
- stop line for missing confirmation

## Procedure
1. load the candidate memory records and their source refs
2. classify each risky record as retain, redact, tombstone, merge, or defer
3. preserve provenance for any proposed removal or masking action
4. avoid silent deletion and require owner approval for destructive memory changes
5. return a pruning proposal with verification and rollback posture

## Contracts
- The skill is explicit-only and must not be invoked as hidden background behavior.
- Titan role, helper/control-plane, runtime implementation, memory, proof, and public-runbook authority stays in owner repositories: aoa-agents, aoa-sdk, abyss-stack, aoa-memo, aoa-evals, and 8Dionysus.
- Receipts, bridge ledgers, console state, replay artifacts, approval records, and memory records are witnesses or candidates, not final owner truth.
- Forge mutation and Delta judgment gates must remain distinct, operator-visible, and receipt-linked.
- Missing source refs, missing approval, missing validation, or unclear owner route must be named as stop conditions.

## Risks and anti-patterns
- treating Titan vocabulary as permission to widen authority
- letting receipt, replay, console, or bridge state replace owner-repo evidence
- auto-approving Forge or Delta because a plan looks plausible
- promoting candidate memory, approvals, replay, or receipts into canon without owner review
- using the skill for an ordinary repo task that has no explicit Titan route

## Verification
- confirm direct skill invocation or explicit Titan service-cohort request is present
- confirm lane, gate, source refs, and owner surface are named when relevant
- confirm Forge or Delta locked or allowed state matches recorded approval evidence
- confirm generated artifacts are marked witness, candidate, derived, or advisory rather than final truth
- confirm next validation, replay, repair, or owner-route follow-up is named before continuation

## Technique traceability
Manifest-backed techniques:
- AOA-T-0103 from `8Dionysus/aoa-techniques` at `3b1d5d623569aa4920b87280d0db0e911d2e29d5` using path `techniques/governance/practice-adoption-lifecycle/adopted-practice-retention-review/TECHNIQUE.md` and sections: Intent, Inputs, Outputs, Core procedure, Contracts, Risks, Validation
- AOA-T-0104 from `8Dionysus/aoa-techniques` at `3b1d5d623569aa4920b87280d0db0e911d2e29d5` using path `techniques/governance/practice-adoption-lifecycle/superseded-practice-obsolescence-route/TECHNIQUE.md` and sections: Intent, Inputs, Outputs, Core procedure, Contracts, Risks, Validation
- AOA-T-0043 from `8Dionysus/aoa-techniques` at `3b1d5d623569aa4920b87280d0db0e911d2e29d5` using path `techniques/instruction/capability-boundary/multi-source-primary-input-provenance/TECHNIQUE.md` and sections: Intent, Inputs, Outputs, Core procedure, Contracts, Risks, Validation

## Adaptation points
- Extract a Titan-specific reusable technique into aoa-techniques only after repeated reviewed evidence exists; do not add pending IDs as placeholders.
- Keep repo-local command examples in owner docs or examples rather than hard-coding them into skill law.
- If a Titan surface graduates from scaffold to reviewed or evaluated, add review evidence before changing status.