Add the skill lifecycle hooks to the taxonomy table

[rocklambros]

What changed

Specification section 5 listed sixteen hooks; the Hooks page lists nineteen. The gap is the entire skill lifecycle set (skillRegister, skillLoad, skillUnload), which reached the schemas, the Hooks page, conformance.md, and the skill AgBOM component type but never the taxonomy table. This adds the three hooks and renumbers the table to nineteen, then guards the set with a test.

Type of change

• Specification change (schema, hooks, events, AgBOM)
• Documentation
• Tooling or CI
• Governance (licensing, security policy, contributor docs)

Specification changes

No normative content changes. The three hooks were already normative in specification/v0.1.0/hooks/ and already documented on the Hooks page. This corrects a summary table that had drifted from them, and it deliberately does not restate the skill rationale, which stays on the Hooks page.

Breaking for implementers? No. Nothing on the wire changes. The practical effect is the reverse of breaking: implementers reading the specification page as canonical were never told skills are governable, so all three adapters in #22 emit no skill hooks at all.

Why it matters

skillRegister is the one point where a Guardian inspects a whole skill definition before any of its actions run. A payload split across a skill's actions is benign at each action and malicious only in composition, so per-action hooks cannot catch it. Omitting the set from the table people treat as the hook list quietly removed that gate from view.

The guard

tests/test_hook_taxonomy.py reads the title field of every schema under specification/v0.1.0/hooks/ as the normative source, then checks it against the section 5 table, the Hooks page overview table, and the hook count the Hooks page states, plus contiguous numbering.

Verified by injection rather than by inspection. Reverting the table fails with Specification section 5 omits hooks the schemas define: skillLoad, skillRegister, skillUnload; changing the stated count fails with Hooks page claims 16 native hooks; the schemas define 19.

This drift happened with a green build from start to finish, so only a check that reads the schemas prevents a repeat.

Not in scope

docs/acs.md carries the same defect ("16 native lifecycle hooks", skill hooks absent), but #21 already rewrites that exact line and names the skill hooks. Fixing it here would conflict with that PR for no gain. If #21 does not land, that line still needs the fix.

The skill hooks also have no OpenTelemetry span or OCSF class in specification/v0.1.0/trace/, so ACS-Trace cannot cover them today. That is a normative decision rather than a doc correction and is filed separately.

Checklist

• Commits are signed off with git commit -s (required by the DCO)
• Prose follows STYLE.md
• uv run mkdocs build --strict passes
• No secrets, tokens, or internal URLs in the diff

Full suite: 198 passed, 1 skipped.