aoa-session-progression-lift

name	aoa-session-progression-lift
description	Turn closed, reviewed session evidence into a bounded multi-axis progression candidate with baseline and attribution limits for aoa-agents review. Do not use on live impressions, without a baseline, to assign one universal score, mutate a role, grant authority, or set routing policy.

aoa-session-progression-lift

Intent

Make observed agent growth legible without turning one success, a mood, or a generated summary into progression truth. The result is an owner-review candidate; aoa-agents keeps progression authority and aoa-evals keeps proof.

Owner-source return

Resolve the canonical aoa-agents root before any owner-relative read:

1. Record <bundle_dir> as the absolute directory containing the SKILL.md actually loaded; never resolve from the task working directory. Initialize one unresolved <source_route> and <owner_root>.
2. In one tool turn containing no other read or command, inspect exactly one same-bundle handle: <bundle_dir>/.aoa-skill-source.json. Await its result. If it is a regular file, set <source_route> to source-handle and require schema aoa_skill_source_receipt_v1 or aoa_skill_source_receipt_v2, this bundle name, owner aoa-agents, version 0.2.5, an existing absolute owner_root, a safe relative source_path, and <owner_root>/<source_path>/SKILL.md. For v2 also require non-empty digest, source_fingerprint, source_fingerprint_scope, and prompt_description_sha256. When capability_graph_hash is present, require it to be a non-empty string and preserve it. If the path exists but is invalid, mismatched, or not a regular file, return blocked_missing_owner_source; do not try another location.
3. Only when that exact same-bundle handle path does not exist, set <source_route> to git and run git -C <bundle_dir> rev-parse --show-toplevel exactly once. Require the returned root's skills/port.manifest.json to declare the expected owner, bundle, and path.
4. In the next tool turn, read only <owner_root>/skills/port.manifest.json; do not include an owner document, bundle reference, evidence read, or unrelated command in that tool turn. Await the manifest result. In the source-handle branch, require the same owner_repo, bundle name, and bundle path as the handle. In the git branch, require aoa-agents, this bundle name, and its actual bundle path. If the manifest read shared a tool batch with an owner document, the serial gate was not observed: return blocked_owner_source_gate_not_observed and do not use either result.
5. Only after the successful manifest result, use a later tool turn to read exactly the two named owner-model documents below. Then owner-source resolution is complete. Do not run or retry the other source branch, including after a later owner-document read fails. Any handle, git, manifest, path, or owner mismatch returns blocked_missing_owner_source and ends this invocation.
6. Never use find, rg --files, parent traversal, sibling scans, workspace conventions, temporary fixtures, .system, or another skill directory to discover a substitute owner root.

Treat handle schema, owner ref, dirty posture, digest, source fingerprint, capability graph hash, and prompt-description hash as install provenance, not authority or current-parity proof. A failed or non-serial resolution is terminal for this invocation; do not synthesize an owner-dependent candidate from the installed copy alone. On success, report <source_route>, <owner_root>, the handle schema and identity dimensions or git action ref, the manifest action ref, and the later owner-model action ref. Read exactly <owner_root>/mechanics/rpg/parts/progression-model/README.md and <owner_root>/mechanics/rpg/parts/progression-model/docs/agent-progression-model.md as the owner model; do not search for substitutes.

Trigger boundary

Use when a closed and explicitly reviewed session contains meaningful mastery, regression, reanchoring, or unlock evidence tied to one agent_id.

Do not use when evidence is live/unreviewed, the agent or baseline posture is unknown, only a single score is wanted, or the requested effect is direct role, rank, authority, routing, quest, or proof mutation.

Inputs and output

• input: one reviewed-progression-evidence packet with agent_id, current session boundary, evidence refs, attribution limits, baseline or explicit baseline state, and optional current progression object
• output: one progression-delta-candidate defined by references/contract.yaml; effect is none until separate owner review

Procedure

1. Read references/contract.yaml and references/lift.md to EOF.
2. Reject live/unbounded evidence and classify every checkpoint, closeout, donor, generated, stale, or neighboring-session item as a hint until the reviewed packet corroborates it.
3. Execute the lift once. Do not apply the candidate to the owner progression object or chain automatically into summon, quest, stats, memo, or proof.

Contracts

• progression is a multi-axis adjunct to agent_id, never a profile rewrite
• comparative movement requires an honest baseline posture
• tools, environment, prior context, and reviewer uncertainty remain visible
• rank and unlocks are reviewable hints; they grant no authority or route rights
• technique lineage is optional provenance, not a runtime dependency

Verification

• trace every moved axis to reviewed refs and state attribution limits
• preserve zero, negative, contested, stale, and no-movement outcomes
• confirm the owner model was read, the candidate was not applied, and the nearest wrong route was rejected