Dense, reference-grade prose that assumes competence (Rust, TOML, crates) and avoids explaining basics; every line carries path/section/trait detail with no padding, so length tracks coverage of multiple surfaces rather than verbosity.

Gives concrete executable guidance throughout — exact file paths, manifest section keys, trait homes, migration function names, and runnable commands like `grep -n 'ID,' ...`, `cargo test -p ironclaw_architecture`, and `pnpm --dir ... test` — pointed at worked examples.

Each surface has a numbered, sequenced workflow with explicit validation: the retired-taxonomy gate as an upfront trip-detector and 'Always finish with cargo test -p ironclaw_architecture' as the machine-reviewed checkpoint, plus error-recovery guidance ('if your change trips that gate, you are re-introducing a deleted model').

Well-sectioned overview per surface with clearly signaled, one-level-deep references to live docs and manifests (e.g. 'overview §4.3', 'Worked example: assets/slack/manifest.toml') and pointers to sibling skills; no nested reference chains.

Lists multiple concrete actions — 'a new extension, a channel surface, model-callable tools, or a shared auth provider' and 'Maps the unified extension model (NEA-25) to the exact manifest sections, crates, seams, and tests' — rather than vague language.

Explicitly answers both: 'what' (maps the extension model to manifest sections, crates, seams, and tests) and 'when' via a clear 'Use when...' clause with multiple triggers.

Natural user phrasing is present, notably 'deciding whether something is "a channel", "an extension", or "a tool"' plus 'adding or changing a Reborn integration', covering the variations a developer would actually say.

Occupies a clear niche (Reborn extension/channel/tool/auth surfaces, NEA-25) with distinct triggers unlikely to fire for unrelated skills.