Agent OS: inject-standards

/inject-standards — deploy the relevant standards into the current context, on demand. This is the consuming half of Agent OS and the wiki’s purest expression of pattern-context-engineering — right context, right time, made an explicit, invocable act. Two modes: auto-suggest (bare /inject-standards analyzes the conversation and recommends from index.yml) and explicit (/inject-standards api/response-format injects a named file). Three steps: context detection (are we coding, building a skill, or planning?) → matching (reads index.yml, matches descriptions against the context, presents 2-5 standards to confirm) → injection.

The injection step surfaces a design choice most frameworks leave implicit — the payload mode:

• Conversation mode loads the full standard content into the window.
• Skill / Plan mode offers a choice between file references (a pointer that stays current as the standard evolves) and embedded snapshots (self-contained, versioned copy).

That reference-vs-embed tradeoff is exactly the guide-layer question of freshness vs self-containment: a reference tracks the living standard but can drift from what was reviewed; an embed freezes what was agreed but goes stale. Injection also surfaces relevant Skills (mentions them) without auto-invoking — reinforcing Agent OS’s Standards (what) vs Skills (how) split, where standards are explicitly invoked and skills auto-detect.

Because it neither authors a lasting artifact nor performs a lifecycle stage — it loads the guide layer into a working context — it carries no canonical stage.

See Also

• agent-os-index-standards — supplies the index.yml this matches against.
• agent-os-discover-standards — authors the standards this injects.
• pattern-context-engineering — the pattern this most fully realizes; compare the harness-side CLAUDE.md/AGENTS.md primitives.
• topic-harness-engineering — guides (feedforward) and the reference-vs-embed tradeoff.