Project Memory Manager
Maintain traceable cross-session project memory with explicit authorization and privacy controls.
The skill restricts inputs to authorized project-local memory, registry projections, approved decisions, and source/date-labelled material. It requires authorization for persistent writes, purge actions, registry proposals, and send-eligibility decisions, and honestly states that Git history, backups, and append-only event payloads may remain. Full marks are not justified because it can modify HOT/WARM/COLD state, indexes, archives, and erasure-related records, while relying on host capabilities and external runtime scripts; static evidence does not prove that isolation, rollback, and data-flow controls work in practice.
Instruction order, authority precedence, stop conditions, path restrictions, conflict handling, and Unknown treatment are comparatively explicit. Points are deducted because initialization, registry, validator, and hook behavior depends on referenced files and scripts not fully contained in this skill, while the supplied static tests do not specifically cover its key paths or provide reproducible failure messages. The static ceiling therefore keeps this below 10.
Triggers, argument hint, read/write scope, completion criteria, canonical-fact boundaries, and erasure limits are reasonably clear; bilingual labeling offers some Chinese-user fit. Points are deducted because initialization requires Claude Code-compatible host behavior, repository-root resolution, and the full repository structure, with standalone single-folder installation unable to initialize. Examples also retain older SEO hot-cache terminology that may cause invocation or operational ambiguity.
Front matter declares name, slug, version, license, compatibility, triggers, and arguments. The document uses progressive sections and linked references, with procedures and examples for the main operations. Points are deducted because maintenance ownership, update path, and changelog governance are not clearly established in the supplied material; SECURITY.md still supports version 17.x while the skill and README state 18.0.0, and several examples do not fully match the current HOT/WARM/COLD model.
The skill defines usable procedures for initialization, querying, consolidation, archival, and erasure, including authorization gates, reporting requirements, and next-skill routing. Points are deducted because static review cannot verify command execution, hook behavior, registry replay, or projection rebuilding. Legacy paths and terminology in examples may require meaningful manual correction, so the score remains within the static ceiling of 7.
Source pointers, event offsets, revisions, hashes, authorization references, and validator requirements provide a useful audit trail. Points are deducted because the evidence is primarily author-provided repository documentation; the supplied CI tests cover architecture and audit-artifact validation, not this skill's key memory-management paths, and no independent execution evidence is provided. Static scoring therefore remains below 5.
- This is a static review only; initialization, querying, archival, consolidation, purge, registry replay, and hook validation were not executed.
- Before use, confirm that the full repository root, referenced files, scripts, and host capabilities are reachable, and resolve the mismatch between skill version 18.0.0 and SECURITY.md support for 17.x.
- Purge is limited to the working tree; Git history, backups, snapshots, and prior event payloads may remain.
- Examples contain older SEO hot-cache paths and terminology and should not be treated as the authoritative current memory model.
What does this skill do, and when should you use it?
memory-management is the project-memory skill in Aaron Marketing Skills’ shared protocol layer. It manages authorized HOT, WARM, and COLD working memory for project context across sessions. It can initialize, review, consolidate, archive, export, or erase working memory while preserving registry ownership and canonical facts. It is a good fit for teams that need durable context without turning cached notes into a second source of truth.
Reads authorized HOT/WARM/COLD memory, registry projections and events, approved decisions, and user-provided or tool-produced artifacts labeled with sources and dates. It can run the repository’s registry-events.py initialization command, check suppression status, query registry offsets, validate or rebuild projections, and perform memory review, consolidation, archival, export, and purge workflows. It writes authorized working notes, indexes, archives, and tombstone or erase events, but never accepts registry proposals or changes canonical facts on behalf of a registry owner. Completed work is reported with authorization status, changed paths or event IDs, preserved conflicts, privacy actions, and the next skill.
- A project lead starts a new agent session and needs to initialize private runtime memory and review current priorities.
- A research team has duplicate working notes and needs to consolidate them without changing registry truth.
- An operations owner needs to review stale WARM files and archive items that have not been updated for more than 90 days.
- A user or data subject submits an explicit erasure request and the team needs to search, suppress communications, and remove authorized working records.
- A multi-skill workflow finishes and needs a session checkpoint so the next session can resume safely.
What are this skill's strengths and limitations?
- Clearly separates HOT/WARM/COLD working memory from canonical registry truth.
- Defines authority ordering, authorization checks, conflict preservation, and privacy-erasure procedures.
- Supports cross-session recovery, note consolidation, archival, and controlled deletion.
- Licensed under Apache-2.0 and intended for Claude Code and compatible Agent Skills hosts.
- Persistent writes, archival, export, and purge are authorization-sensitive and may require confirmation.
- Requires local filesystem access, a shell, Python 3, and Git; a standalone one-folder install cannot initialize or claim registry state.
- Claude Code hook integration may require adaptation on other hosts.
- The supplied source does not document an independent test suite or broad platform coverage for this individual skill.
How do you install this skill?
Install the repository bundle with /plugin marketplace add aaron-he-zhu/aaron-marketing-skills followed by /plugin install aaron-marketing@aaron in Claude Code. Other Agent Skills-compatible hosts can use npx skills add aaron-he-zhu/aaron-marketing-skills, or clone it with git clone https://github.com/aaron-he-zhu/aaron-marketing-skills. The skill is located at protocol/memory-management/SKILL.md. The README does not document a dedicated single-skill installation command for this skill.
How do you use this skill?
Ask the compatible host for a concrete operation such as “initialize project runtime memory,” “review saved findings,” “archive WARM files not updated in 90 days,” or “purge subject-7f42 under this confirmed erasure request.” Read-only lookup, verification, and dry-run planning do not require new write consent; persistent changes generally require explicit authorization. Initialization uses python3 "$AARON_SKILLS_ROOT/scripts/registry-events.py" init and checks the runtime directories and .gitignore state.