Memory Recall

Use this skill when the current task may depend on prior work.

Typical triggers:

• existing project
• user says "we discussed this before"
• architecture or product decision may already exist
• task failed previously
• reviewer needs project conventions
• an autonomous mission resumes after interruption

Goal

Retrieve the smallest useful historical context.

Do NOT:

• load whole Obsidian vault
• load whole conversation history
• retrieve every project note
• expose secrets or excluded memory roots

Workflow

1. Rewrite the current need as a precise retrieval query.
2. Call memory.search.
3. Inspect the top summaries/snippets.
4. If necessary call memory.expand only on the strongest few results.
5. Prefer:
   • newer explicit decisions
   • project-specific evidence
   • verified outcomes
     over generic or older notes.
6. Detect conflicts between memories.
7. Return a compact context pack with source references.
8. Inject only the context required for the next reasoning step.

Default budget

Target:

• 6–12 initial retrieval candidates
• 3–6 final evidence items
• <= ~7000 memory tokens unless mission policy explicitly allows more

Conflict handling

If two memories disagree:

• do not silently choose one
• compare timestamps/provenance
• prefer explicit superseding decisions
• state uncertainty if unresolved

Output

Provide:

• relevant decisions
• useful lessons
• unresolved conflicts
• source references
• confidence
• what was deliberately omitted as irrelevant

Memory is evidence, not infallible truth.