mnemos: Instructions file for Claude Code

Memory Integration

Use Mnemos automatically during this workspace session.

Retrieval

At the start of a new task or session, call mnemos_context once before planning, coding, or delegating to a subagent.

• Use the current task, bug, feature, or subsystem as the query
• Use the current project scope
• Use a token budget around 2000
• Do not call repeatedly unless the task changes substantially

Before Delegating To Subagents

Claude subagents may not receive session-start hook context or parent steering context.

Before using the Task tool or delegating to any subagent:

1. Call mnemos_context with the task/feature/bug as query.
2. Read the returned context.
3. Include the relevant Mnemos context or memory IDs in the subagent prompt.
4. Tell the subagent to call mnemos_search if it needs deeper project memory.

Example:

Task: Review payment collection bug.
Mnemos context: payment collections use current-shift scope; hotel history requires explicit hotel scope.
Use these constraints while reviewing backend handlers and frontend list filters.

Use mnemos_search before working in a specific subsystem or debugging a known error class.

When to Store

Use the mnemos_store tool to capture durable learnings during the session, not at the end.

Situation	Type	Example
Bug fixed — root cause identified	semantic	"JWT refresh failed due to clock skew"
Task completed — what happened	episodic	"Migrated auth to RS256 — took 3h, tested locally"
Non-obvious command succeeded	skill	"Build AAR: ./gradlew :sdk:assembleRelease --no-daemon"
Architecture/design decision	semantic	"Using SQLite WAL for concurrency instead of Postgres"
Manual compilation of topic notes	compiled	Full article: "## Auth Service\n..."

Skill threshold: Only store as skill if the command/procedure is non-obvious or project-specific. Don't store git status, ls, cd. Do store project-specific commands with flags, paths, or ordering requirements.