Aoa Session Self Repair

aoa-session-self-repair

Intent

Use this skill to author a bounded REPAIR_PACKET.

The repair packet should be small, explicit, reversible where needed, and aligned with checkpoint posture before any important system surface is changed. It should also say whether the repair is only proposed, prepared, executing, verified, blocked, or handed off.

Trigger boundary

Use this skill when:

• a reviewed diagnosis already exists
• the next honest move is a bounded repair plan or repair-ready packet
• the route may change skill, playbook, agent, eval, or memo surfaces and needs explicit checkpoint posture
• the route may need prerequisite repair before later automation becomes honest
• repair can still be kept inside one bounded execution unit
• a prepared repair packet or receipt must not be mistaken for executed or verified repair

Do not use this skill when:

• there is no reviewed diagnosis yet
• the repair is actually a large scenario rollout better owned by a playbook
• the route is trying to bypass approval, rollback, or health-check posture
• the request is vague self-improvement rhetoric with no bounded target
• the route only needs a diagnosis packet, route fork, automation-readiness scan, or final quest-promotion verdict

Inputs

• reviewed diagnosis packet
• owner-layer candidates
• risk and approval posture
• known validation surfaces
• current rollback anchors if any
• current execution state, if a repair packet already exists
• evidence showing whether validation actually ran or is only planned

Outputs

• REPAIR_PACKET with target owner repo, smallest diff shape, approval need, rollback marker, health check, and improvement-log stub
• explicit repair execution posture such as proposed, prepared, executing, verified, blocked, or handoff_required
• optional repair quest when execution should remain deferred
• optional automation-readiness prerequisite packet when the real need is to

stabilize a route before later automation scanning or seeding

• explicit stop conditions and escalation points
• one REPAIR_CYCLE_RECEIPT using references/stats-event-envelope.md and

references/repair-cycle-receipt-schema.yaml

• one CORE_SKILL_APPLICATION_RECEIPT using

references/core-skill-application-receipt-schema.yaml

Procedure

1. start from the reviewed diagnosis rather than from general aspiration
2. choose the smallest honest repair shape
3. name the primary owner repo and target artifact class
4. state the repair execution posture before writing any outcome language: proposed, prepared, executing, verified, blocked, or handoff_required
5. record checkpoint posture: constitution or policy check, approval gate, rollback marker, post-change health check, bounded iteration limit, improvement log
6. if validation has not actually run, keep health checks as planned checks and set execution posture below verified
7. if the target route was blocked automation, emit the smallest prerequisite repair that would make later automation classification more honest
8. define validation and stop conditions
9. emit a repair quest instead of mutating immediately when risk or approval posture requires it
10. emit one REPAIR_CYCLE_RECEIPT when the repair packet or repair-quest handoff closes, keeping the receipt smaller than the packet
11. when the finish path closes, emit one CORE_SKILL_APPLICATION_RECEIPT

that points back to the bounded repair receipt and records one finished kernel-skill application

Contracts

• self-repair is not free self-modification
• important surface changes must pass checkpoint posture
• repair packets stay bounded and reviewable
• prepared repair packets are not executed repairs, and executed repairs are not verified repairs unless validation evidence is present
• role law changes route to aoa-agents
• proof-law changes route to aoa-evals
• scenario-scale repair routes to aoa-playbooks
• repair does not smuggle live automation authority into the packet
• repair-cycle receipts stay descriptive and append-only
• generic core receipts stay subordinate to the repair receipt and do not

replace checkpoint or repair meaning

• receipt corrections use supersedes rather than silent overwrite

Risks and anti-patterns

• silent doctrine edits
• approval bypass
• retry loops disguised as repair
• using repair to hide broader governance debt
• changing too many surfaces at once
• letting a repair receipt pretend the repair is already verified when it is only planned
• using resolved, prepared, or finished without saying what was actually executed and verified

Verification

• confirm diagnosis exists and is cited
• confirm the chosen repair is the smallest honest shape
• confirm checkpoint fields are present
• confirm repair execution posture is explicit and does not overclaim validation
• confirm validation and rollback posture are named
• confirm escalation route exists if the repair widens
• confirm any emitted receipt cites diagnosis and validation refs without duplicating the whole packet
• confirm any emitted CORE_SKILL_APPLICATION_RECEIPT points to the repair

detail receipt and stays finish-only

Technique traceability

Manifest-backed techniques:

• AOA-T-0082 from 8Dionysus/aoa-techniques at 3b1d5d623569aa4920b87280d0db0e911d2e29d5 using path techniques/recovery/diagnosis-repair/repair-shape-from-diagnosis/TECHNIQUE.md and sections: Intent, Inputs, Outputs, Core procedure, Validation
• AOA-T-0083 from 8Dionysus/aoa-techniques at 3b1d5d623569aa4920b87280d0db0e911d2e29d5 using path techniques/recovery/diagnosis-repair/checkpoint-bound-self-repair/TECHNIQUE.md and sections: Outputs, Contracts, Risks, Validation

Adaptation points

Project overlays may add:

• local approval classes
• repo-specific rollback anchors
• bounded repair templates

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: 8Dionysus
• Source: 8Dionysus/aoa-skills
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.