Engram tools have clear, domain-specific descriptions (avg 95 chars) and well-structured input schemas with proper types and constraints. However, critical gaps undermine production readiness: (1) No output schemas documented, LLMs cannot predict return structure or plan downstream calls; (2) Parameter descriptions lack actionable constraints (e.g., 'scene' and 'content' are vague strings with no format guidance); (3) No error handling guidance, tools return bare errors without recovery hints; (4) Missing idempotency/retry semantics despite WRITE operations; (5) No permission/scope declarations despite sensitive state mutations. Tool names follow verb_noun pattern well (summon, wake, observe, release, outcome, fold_status, fold_revert), but lack context about prerequisites and dependencies between operations.

• Document output schemas for all tools. Example: engram_summon should return {accompaniment_id: string, journal_id: string, created_at: ISO8601}. This enables LLMs to chain calls and extract required IDs.
• Add format/constraint descriptions to string parameters. E.g., 'scene: A brief context string (max 500 chars, no newlines)' and 'content: The exact message text (max 10000 chars, supports markdown)'. This prevents invalid input and guides LLM reasoning.
• Add error recovery hints to tool descriptions. E.g., 'If accompaniment_id is invalid, call engram_summon() to create a new accompaniment. If the Engram is not found, check engram_fold_status() to verify it exists.'
• Declare idempotency guarantees. E.g., 'engram_observe is idempotent: calling it twice with the same accompaniment_id and content appends only once.' This guides agent retry logic.
• Add permission/scope declarations. E.g., 'Requires: write:engram, read:journal. Scope: Engram state mutation.' This enables least-privilege agent configs and audit trails.