Project Memory: Explicit Context for Codex Agents

Where this fits

You’ve been away from a codebase for a week. You open your editor, ask Codex to continue work on the payments API, and it has no idea what you decided about the database schema or the auth flow. You waste minutes re-explaining context that should have been persisted. This skill solves that by giving agents a simple, explicit memory store—no hooks, no background capture, just a file you control.

Why agents benefit

• Durable decisions: Save one-line notes like “Use PostgreSQL for payments” that survive across sessions and agent restarts.
• Explicit control: The agent only writes memory when you ask—no silent recording of tool traffic or conversation history.
• Collision detection: Before merging notes, the skill verifies the canonical project path, preventing cross-project contamination or moved-directory mixups.
• Recoverable archives: Old decisions move to an archive directory instead of being deleted, so you can audit or restore them later.
• Lightweight and portable: Stores plain Markdown files under ~/.codex/memory/, so you can inspect or sync them with your own tooling.

Practical scenarios

Resuming after a break

You return to a feature branch after a long weekend. You say, “Resume prior work on the checkout flow.” The agent reads the project memory, sees the note about Stripe integration decisions, and picks up without re-asking for context.

Handoff between agents

You run one agent for refactoring and another for testing. After the refactor, you say, “Save a handoff for the testing agent.” The memory file gets a concise summary of changed modules and known risks, so the next agent starts informed.

Recording architectural decisions

During a code review, you decide to switch from REST to GraphQL. You say, “Remember that we’re migrating the public API to GraphQL.” The agent writes that decision with the project path and timestamp, preventing future drift.

Add it to your agent workflow

Install the skill with:

npx skills add aAAaqwq/AGI-Super-Team --skill project-memory

Then, in a Codex session, simply say:

Remember that we’re using PostgreSQL for the new payments service.

The skill creates or updates ~/.codex/memory/projects/payments-api-<hash>/memory.md with that note. To recall, ask “What did we decide about payments?” and the agent reads the file.

Compare before adopting

Before making this a permanent part of your workflow, compare it against alternatives like full-conversation logging or vector-based memory. Check how often the repository is updated (last commit, open issues), whether the storage format fits your existing backup or sync habits, and if the collision detection logic is robust enough for your monorepo structure. Also consider if you prefer a more automated approach—this skill is deliberately manual, which may be a feature or a limitation depending on your trust level.

Why it is worth tracking

With 91 stars and an active author, this skill is a pragmatic, minimal solution to a common agent pain point: context loss. Its explicit, file-based design makes it auditable and hackable, and the collision handling shows attention to real-world edge cases. Keep an eye on how it evolves—if the author adds optional hooks or a richer query syntax, it could become a default choice. You can also discover this skill and others on OpenAgentSkill at https://www.openagentskill.com/skills/aaaaqwq-project-memory. Evaluate it when you want predictable, user-controlled memory without the overhead of a full agent memory framework.