recall - SKILL.md Agent Skill

name: recall description: Search past knowledge in the sayou workspace user_invocable: true triggers: - "what did we decide about" - "do we have research on" - "find notes about" - "what do we know about" - "search workspace for" - "recall"

/recall — Search Past Knowledge

Search the sayou workspace for past decisions, research, notes, and other saved knowledge.

Instructions

Use a multi-step search strategy to find the most relevant results:

Step 1: Structured search (fastest)

Use workspace_search with the user's query. This searches file paths and frontmatter text.

workspace_search(query="auth architecture", limit=10)

If the user mentions a specific type or status, add filters:

workspace_search(filters={"type": "decision"}, query="database")
workspace_search(filters={"status": "approved"})

Step 2: Content search (if Step 1 is insufficient)

Use workspace_grep to search inside file contents:

workspace_grep(query="PostgreSQL", context_lines=3)

Step 3: Semantic search (if Steps 1-2 miss)

Use workspace_semantic_search for meaning-based retrieval when exact keywords don't match:

workspace_semantic_search(query="why we chose that database", top_k=5)

Note: This only works if the workspace has embeddings enabled (SAYOU_EMBEDDING_PROVIDER set).

Step 4: Read and present

For each relevant file found, use workspace_read to get the full content. Present the key information clearly, citing which file it came from and when it was last updated.

Output Format

Present results with clear attribution:

Found in decisions/database-selection.md (v2, updated 3d ago):

  Decision: PostgreSQL over MySQL
  Status: approved
  Reasoning: Better JSON support, superior FTS, row-level security

If multiple files match, show a summary list first, then offer to read specific ones in detail.

Tips

Start with the broadest reasonable search, then narrow down
If nothing is found, tell the user — don't make up information
Mention the file path so the user can reference or update it later
If the workspace is empty, suggest using /save to start building knowledge
For time-based queries ("what did we do last week"), use workspace_search with date filters or check activity/ logs