# Aoa Memo Writeback > Use only before any memo artifact exists to decide whether live-session, closeout, landed-work, PR, diff, or review evidence yields one guarded memo candidate, export, or no-write result. If a concrete memo candidate, export, object, corpus, lifecycle, or read-model target already exists, use aoa-memo. Do not use for eval mining, raw-session maintenance, generic harvest, recall, durable memory, or proof. - Skill: `8dionysus/aoa-memo-writeback` (Agent Skill, multi-file: 4 files) - Install (CLI): `npx skillmds@latest add 8dionysus/aoa-memo-writeback` - Raw SKILL.md: https://api.skillmd.com/api/skills/8dionysus/aoa-memo-writeback/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Marketing & Growth - Author: 8Dionysus (https://skillmd.com/u/8dionysus) - Updated: 2026-09-10 - Page: https://skillmd.com/skills/8dionysus/aoa-memo-writeback --- # aoa-memo-writeback ## Intent Preserve the first writeback judgment that the previous consolidation removed: notice when live or closing work produced a bounded lesson worth remembering, decide its real owner, and create or route its first guarded artifact without turning session history into memory truth. ## Trigger boundary Use this skill when: - a live session, closeout, PR, diff, review, release, or recurring correction may contain a durable owner lesson not captured by the final artifact alone, and no concrete memo candidate, export, or memory object exists yet - the decision is still whether to write a local candidate, prepare an export, record route-only debt, request owner review, or write nothing - the caller explicitly authorizes creation of the first guarded local candidate from inspectable work evidence Do not use this skill when: - the task is eval-trigger/session-case mining, `.aoa` preservation/index repair, generic session harvest, review of an existing candidate/export/ quarantine packet/object, corpus recall or lifecycle work, direct durable `aoa-memo` landing, or proof/scoring work ## Inputs - likely owner and stronger owner; task/closeout/landed-work context; inspectable session, source, PR, review, or receipt refs; local memo-port status; privacy, freshness, review posture, and confirmation that no concrete memo artifact already represents this writeback ## Outputs - one `memo_writeback_decision`: `write_candidate`, `prepare_export`, `no_writeback_needed`, `route_only_debt`, `needs_owner_review`, or `blocked` or `owner_handoff` - when authorized, one guarded owner-local candidate or reviewed export packet; otherwise a no-write stop line or route-only debt with inspectable refs ## Procedure 1. Before reading this bundle, inspect only the request. If it already names or supplies a concrete candidate, export, quarantine packet, memory object, corpus identity, lifecycle target, or memory read-model target, do not load this skill. Select `aoa-memo` directly. 2. After selection, the first tool turn must read only this bundle's `SKILL.md` from the exact absolute package path observed in the host catalog or returned by a verified owner resolver. Use that loaded package path as the source root for the references below. Catalog syntax such as `file:` is only a locator presentation; a normalized absolute path is sufficient. Do not reconstruct the path from a presumed skill root, `.system`, `$CODEX_HOME`, the workspace, or the skill name. Do not combine the read with `pwd`, `AGENTS`, workspace inspection, or another command. An absent, malformed, non-absolute, or ambiguous catalog/resolver identity is `blocked_package_path_not_observed` before any other package or workspace action. If one constructed package path already failed with a pure not-found result before this body loaded and no workspace, search, or other action occurred, read the exact host locator or verified resolver result next and set `package_path_recovered: true`. This is a degraded one-call recovery, not a clean package entry; report it in the result and do not claim equivalent routing cost. Any second miss, search, directory probe, workspace read, or absent/failed locator or ambiguous resolver result is `blocked_package_path_not_observed`. 3. Read `references/contract.yaml`, then `references/writeback.md`, from that same loaded package path. Do not inspect the task workspace before both are loaded. 4. Check whether the inspected evidence unexpectedly supplies a concrete candidate, export, quarantine packet, memory object, corpus identity, or lifecycle target. If it does, return `owner_handoff` to `aoa-memo` without opening owner-corpus surfaces or creating another artifact. 5. Name three distinct owners before judging whether the material belongs in memo: - `origin_owner` owns the source claim and local candidate; - `memory_intake_owner` is `aoa-memo` for reviewed intake and durable-memory lifecycle only; - `procedure_owner` is `aoa-skills` and is provenance, never the stronger owner of the material. If the caller requests `owner` and `stronger_owner`, return `origin_owner` and `aoa-memo` respectively, while stating that `aoa-memo` cannot override origin source truth. 6. Use session evidence when the meaning lived in the interaction; reconcile it with source/PR/review evidence without letting either silently replace the other. 7. Choose exactly one decision and honor its effect boundary. Candidate writing requires an existing admitted local `memo/` port and explicit authorization. Before choosing between a local candidate/review route and `route_only_debt`, read the exact conventional `memo/PORT.yaml` once. A closeout statement that a port is admitted is evidence to check, not port authority. Use its `owner_repo` as the repository-level `origin_owner` while keeping any missing exact material-source path visible. 8. Stop before durable memory landing unless the active task explicitly enters the `aoa-memo` reviewed-corpus owner route. ## Contracts - `.aoa` is evidence and routing context, never reviewed memory truth - owner-local candidate, export packet, reviewed intake, and durable memory are separate lifecycle states and effects - one candidate carries one bounded memory question with source and evidence refs; it is not a session recap - an existing candidate, export, object, or corpus question belongs to `aoa-memo`; writeback never creates a second representation to avoid review - missing owner ports produce route-only debt, not invented directories - technique records may explain provenance but are not runtime dependencies ## Risks and anti-patterns - treating every commit, progress note, mood, or interesting thought as memory - using a clean diff summary instead of the session evidence that explains why a correction matters - smuggling proof, routing, runtime, role, playbook, KAG, ToS, source doctrine, secrets, or private raw transcript detail into memo ownership - turning `reviewed_write` or an MCP dry run into durable authority ## Verification - confirm owner/stronger-owner, bounded memory question, source and evidence refs, freshness, privacy, local-port status, chosen decision, and effect - confirm package identity and the exact loaded source path or verified resolver used - confirm candidate guardrails and review posture; inspect any packet manually - after an authorized candidate write, use the file-change result and one exact read of that candidate as the complete effect/artifact check; do not run `git status`, `pwd`, `stat`, directory inspection, or another command merely to restate the write - report `package_path_recovered` when the one-call not-found recovery occurred; keep its extra call visible instead of rewriting it as clean package entry - report what was written, what stayed local, what was not promoted, validation used, and the next owner route ## Adaptation points Owners supply memory-worthy trigger classes, local port schemas/builders, privacy rules, export policy, reviewed-intake route, and closeout conventions.