polo871209

Read before writing or editing any agent-facing instructions — an agent skill (SKILL.md and whatever it links to), a tool/function description, or an extension's injected context. Covers discoverability, information hierarchy, and token-lean design.

Strip dead code and useless comments from files touched this session, per house rules.

Standards review of the diff since a fixed point; Spec axis added only when a spec exists.

Compact the current conversation into a handoff document for another agent to pick up.

Reference for writing/editing LLM system prompts and tool descriptions in this repo. Compresses prose to caveman style, preserves intentions and code.