| name | aoa-session-memory-global-route |
| description | Use in any Codex session when the user mentions `.aoa`, session memory, Codex transcripts, compaction, prior session rehydration, hook failures, or AoA session-memory validation. |
| license | Apache-2.0 |
| metadata | {"aoa_scope":"session-memory","aoa_invocation_mode":"global-router"} |
aoa-session-memory-global-route
Use this as the top-level router for the AoA session-memory bundle.
Source Root
Resolve three logical roots before acting:
<workspace-root>: the selected project workspace;
<aoa-root>: the active installed session-memory root;
<portable-source-root>: the standalone owner source when source/export
work is required.
Concrete paths are runtime bindings, not portable skill identity.
Trigger Boundary
Use this skill in any Codex session when the task touches:
.aoa session memory
- Codex raw transcript JSONL
- context compaction or compaction intervals
- prior-session resume, rehydration, or session archive lookup
- AoA hooks for
SessionStart, UserPromptSubmit, PreCompact,
PostCompact, or Stop
raw_unavailable incidents
stress-pass, audit, doctor, codex-hooks-status, or
codex-compact-probe
- event taxonomy, classifier, generated segment indexes, or
reindex-sessions
- portable SQLite search,
search-index, aoa-search,
search-provider-status, optional host provider gates, or retrieval
freshness
- retrieval packets,
retrieve, retrieval-packet, or long-session
continuation recipes
- live route-quality checks,
live-scenario-audit, or
live-scenario-corpus
- naming readiness,
SESSION_NAMES.md, sessions/INDEX.md, or
naming-readiness
- mass session naming,
naming-wave, semantic session-name review plans, or
naming quality audit
- phase discovery,
phase-discovery, review-phase-name, or session naming
candidate layers
- historical Codex session import from
<codex-history-root>
- first-wave batch distillation or historical session review queues
- preparing or validating the portable
aoa-session-memory bundle
Procedure
- Read
<aoa-root>/AGENTS.md and classify the task before loading
deeper context.
- Read only the owning surface:
DESIGN.md for architecture or boundaries,
DESIGN.AGENTS.md for query/access behavior, the relevant PIPELINE.md
section for operations, INSTALL.md for portability, NAMING.md for names,
or READINESS.md for proof posture.
- Read
references/capability-router.md, verify
its source hash against the generated graph when composition matters, and
choose the smallest applicable bundle.
- Use
<aoa-root>/scripts/aoa_session_memory.py for commands.
- Keep historical raw/session material intact unless the user explicitly asks
for a repair.
- If the task changes portable behavior, export to
<portable-source-root> and validate both source and
standalone surfaces.
- If the user-level router itself is missing or stale, run
install-user-skill from the active install root instead of hand-writing a
symlink.
Skill Routing
Use the generated router card for exact package names, positive and negative
applicability, visibility, version, and fingerprint. Its stable branches are:
use.route for top-level selection;
use.query for evidence, search, and rehydration;
stewardship.capture, .project, .curate, .name, and .assure;
adapters.codex for Codex hooks, transcripts, and compaction.
For more than one capability, use the full generated graph and task-local DAG
planner. Do not select a set from topical similarity alone.
Verification
- The chosen capability's positive trigger matches and its negative trigger
does not.
- The selected package version/fingerprint matches the current generated
router and installed runtime receipt when available.
- Required inputs, permissions, effects, and verifier are known before action.
- The prompt-visible set remains the two admitted routers unless a separate
routing eval and owner decision changes it.
Stop Line
Do not replace raw evidence with summaries. Use indexes and diagnostics first,
then open raw only for exact verification or repair.