Compatibility entrypoint for runtimes that auto-detect AGENTS.md.
CRITICAL: You MUST read and follow KNOWNS.md in the repository root before doing any work. It is the canonical source of truth for all agent behavior in this project.
- Knowns is the repository memory layer for humans and the AI-friendly working layer for agents.
- The source of truth for repo-level agent guidance is
KNOWNS.md. - Read
KNOWNS.mdfirst whenever the runtime supports reading repository files. - Load behavior, memory policy, and workflow rules from
KNOWNS.md; treat this file only as a compatibility entrypoint. - If this file and
KNOWNS.mddiffer, followKNOWNS.md.
- Use Knowns as the canonical system for tasks, docs, templates, and workflow state.
- Never manually edit Knowns-managed task or doc markdown.
- Search first, then read only relevant docs and code.
- Use
searchfor discovery; use MCPretrievetool when a workflow needs structured context with citations. Fall back to CLIknowns retrieveif MCP is unavailable. - For code context retrieval, prefer MCP tools over CLI: use
code({ action: "search" })first, thencode({ action: "symbols" }), thencode({ action: "deps" }). Treat CLIknowns code ...as fallback for manual inspection or debugging. - Plan before implementation unless the user explicitly overrides that workflow.
- Validate before considering work complete.
- Use memory tools:
memory({ action: "list" })at session start,memory({ action: "add" })after tasks for reusable knowledge. - Proactively capture durable memory based on
KNOWNS.mdmemory rules; do not wait for an explicit user instruction to save memory when scope and durability are clear.
knowns doc list --plain # List docs
knowns task list --plain # List tasks
knowns task <id> --plain # View task
knowns doc "<path>" --plain --smart # View doc
knowns search "query" --plain # Search docs/tasks
knowns retrieve "query" --json # Retrieve structured context pack (CLI fallback)