proactive-memory
Skill4 starsChanged 7 months agoMITView on GitHub
Proactively remember project conventions, architecture decisions, gotchas, and corrections during normal coding work. When an agent: (1) Discovers a naming convention or code pattern, (2) Learns why something is built a certain way, (3) Encounters a non-obvious gotcha or workaround, (4) Gets corrected by the user on approach or style. Captures valuable context so future sessions don't start from scratch.
In this repository2 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- session-contextSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 46
When the user corrects your approach — remember it so you don't repeat the mistake.
- Very specific: line 66
When you identify something worth capturing, call the remember tool:
- Distinctive: line 81
Keep it short. Don't interrupt the flow of work to celebrate remembering something.
- Distinctive: line 104
Not everything needs to be remembered.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Proactive Memory — Remember as You Work
Keywords
remember this, save convention, note this pattern, don't forget, save this decision, remember for next time, keep this in mind, convention, architecture decision, gotcha, correction
Overview
During normal coding work, capture valuable project knowledge into Tages so it persists across sessions. This skill guides when and what to remember — the goal is that the next session starts with full context instead of amnesia.
Use this skill when: You discover something worth remembering during coding, or the user explicitly asks you to remember something.
What to Remember
Conventions
Naming patterns, code style, file organization, import conventions, test patterns.
Examples:
- "Components in this project use PascalCase with a
.component.tsxsuffix" - "All API routes validate input with Zod schemas before processing"
- "Tests use factories from
tests/factories/instead of inline fixtures"
Architecture Decisions
Why something is built a certain way, trade-offs that were made, structural choices.
Examples:
- "Auth uses Supabase PKCE flow with a /auth/cli-token workaround for CLI OAuth"
- "SQLite is the primary store for sub-10ms reads; Supabase syncs in background"
- "Dashboard deploys manually via
vercel --prod, not auto-deploy from GitHub"
Gotchas
Non-obvious behavior, common mistakes, workarounds, things that look wrong but are intentional.
Examples:
- "Supabase
.from().insert()returns PromiseLike not Promise — wrap with Promise.resolve() for .catch()" - "The ESM/CJS boundary between CLI and server requires createRequire + directory walk"
- "Migration params need explicit
::uuidcast for uuid columns"
Corrections
When the user corrects your approach — remember it so you don't repeat the mistake.
Examples:
- User: "No, we don't mock the database in these tests" → Remember: integration tests use real DB
- User: "Use the existing Button component, don't create a new one" → Remember: reuse shared components from
src/components/ui/
What NOT to Remember
- Ephemeral task details — current bug description, in-progress work status
- File contents or code — derivable from reading the codebase
- Debugging steps — not reusable across sessions
- Anything in CLAUDE.md or README — already persisted and loaded automatically
- Obvious language/framework behavior — only project-specific knowledge
How to Remember
When you identify something worth capturing, call the remember tool:
remember(
key: "descriptive-short-name",
value: "The thing to remember — clear, specific, actionable",
category: "convention" | "architecture" | "decision" | "gotcha",
context: "Why this matters or when it applies"
)Then briefly inform the user:
Remembered: [key]
Keep it short. Don't interrupt the flow of work to celebrate remembering something.
When to Remember
Do remember when:
- You discover a pattern after reading multiple files ("ah, all services follow this structure")
- The user corrects you ("no, we do X not Y")
- You hit a gotcha that wasted time ("this API doesn't work the way you'd expect")
- The user explicitly says "remember this" or "note this for next time"
- You make an architecture decision together ("let's use approach B because...")
Don't remember when:
- You're unsure if it's a convention or a one-off
- The information is already stored (check with
recallfirst if uncertain) - It's generic knowledge, not project-specific
- The user is brainstorming and hasn't decided yet
Judgment Calls
Not everything needs to be remembered. Use judgment:
- High value: Patterns that affect multiple files, decisions that took discussion to reach, gotchas that caused debugging time
- Low value: Minor preferences, single-file details, things that are obvious from reading the code
- When uncertain: Lean toward remembering. It's easier to forget (clean up) later than to re-discover something lost.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

