agentsclimarketplace

Aoa memo writeback

Skill 8Dionysus/aoa-skills/skills/core/session-growth/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.From its SKILL.md

Install
npx -y skills add 8Dionysus/aoa-skills --skill aoa-memo-writeback

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

7.0 KB, ~1.5k tokens by cl100k_base, as published. Nobody here has run it

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 by copying the absolute (file: ...) source locator appended by the host to this skill's catalog entry. 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.

    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 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 literal locator is blocked_package_path_not_observed.

  3. Read references/contract.yaml, then references/writeback.md, from that same bundle. 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 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.

What ships with it: 3 files

9.3 KB alongside SKILL.md

agents/

references/

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.