Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@cmd/root_test.go`:
- Around line 78-85: The current test only verifies a static footer URL and
misses the actual behavior change in root help generation. Update or add tests
around the root command help flow to cover the new grouping logic and the lazy
PrepareDomainHelp and PrepareMethodHelp paths in cmd/root.go, so regressions in
those branches are exercised instead of only checking rootUsageTemplate. Use the
existing TestRootLong_AgentSkillsLinkTargetsReadmeSection as a starting point,
but assert the command/help behavior that changes when the new branches run.

In `@cmd/root.go`:
- Around line 571-584: The command grouping logic in root.Commands() is missing
completion, so it falls into “Additional Commands” instead of the management
section. Update the classifier in cmd/root.go to treat c.Name() == "completion"
the same as the other management commands by adding it to the management mapping
used before setting c.GroupID.

In `@cmd/service/affordance.go`:
- Around line 24-65: Add tests covering the new help-generation branches so
`PrepareDomainHelp` and `PrepareMethodHelp` don’t regress silently. Exercise the
`PrepareDomainHelp` path for the shortcut/resource routing text and the optional
domain-guide line when `skillFS` contains the matching `SKILL.md`, and add
coverage for the high-risk/help workflow branches in `PrepareMethodHelp` by
asserting the expected help output changes. Use the existing `PrepareDomainHelp`
and `PrepareMethodHelp` symbols to locate the logic and verify both the gated
and non-gated cases.

In `@cmd/service/paramhelp.go`:
- Around line 33-36: The boolean help text in paramhelp.go is inaccurate because
it describes bool flags as presence-only. Update the logic in the help text
generation for CanonicalType() == "boolean" so it explains that bool flags can
be set explicitly to true or false (for example via the flag handling around
facts append), instead of implying omission is the only way to get false. Keep
the wording aligned with the existing paramhelp.go formatting and the symbol
names CanonicalType and facts.

In `@content_embed.go`:
- Around line 30-40: The warn-and-continue embed fallback in init() for
embeddedContentFS is not directly testable, and the fs.Sub failure paths for
skills and affordance are uncovered. Extract the subtree wiring into a small
helper (for example around cmd.SetEmbeddedSkillContent and affordance.SetSource)
and keep init() as a thin caller, so tests can exercise both successful fs.Sub
results and warning-only failures. Add tests for the helper to verify the
warning branches and the happy path behavior.

In `@internal/affordance/affordance_test.go`:
- Around line 31-33: The TestFor setup currently mutates the package-global
affordance source via SetSource without restoring it, which can leak state into
other tests. Update the TestFor test to save the previous source before calling
SetSource and restore it in t.Cleanup so each test leaves the global state
unchanged; use the SetSource helper and the existing test function name TestFor
to locate the setup.

In `@internal/affordance/mdparse.go`:
- Around line 141-160: Paragraph-style text in non-Examples sections is being
stored in pending but never emitted, so guidance under sections like Tips,
Prerequisites, or custom headings gets dropped. Update the section parsing in
mdparse.go around the fenced-block and pending handling so non-bullet, non-fence
lines are preserved as section content, and ensure pending is flushed into
sec.items or another emitted structure when appropriate instead of only being
consumed by meta.AffordanceCase creation. Keep the behavior consistent with the
existing parsing flow in the section parser and the linkToBacktick / sec.items
accumulation path.

In `@internal/schema/assembler_test.go`:
- Around line 513-516: The test in assembler_test.go mutates package-global
affordance source via affordance.SetSource and does not restore it, leaving
later tests dependent on execution order. Update the test to save the original
source before installing the fstest.MapFS fixture, then restore it after the
test using defer so the default source is preserved for any other tests that use
affordance or assembler_test helpers.

---

Outside diff comments:
In `@internal/meta/affordance_test.go`:
- Around line 36-59: The populated-case test in ParsedAffordance currently
verifies UseWhen, Tips, Examples, and Related, but it does not assert the newly
added AvoidWhen and Prerequisites fields, so a regression in those parsed keys
could slip through. Update the affordance test around Method.ParsedAffordance to
check that the parsed affordance includes the expected avoid_when and
prerequisites values from the raw fixture, alongside the existing assertions.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@affordance/mail.md`:
- Around line 42-52: The EML guidance in the mailbox draft docs uses the wrong
RFC number and should reference the email message format standard instead.
Update the text in the user_mailbox.drafts create section, and any related
draft/send/reply guidance it points to, so every occurrence of RFC 5822 becomes
RFC 5322 in the mail documentation. Keep the wording otherwise unchanged and use
the existing user_mailbox.drafts create labels to locate all matching
references.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Duplicate comments:
In `@affordance/mail.md`:
- Line 43: The EML guidance uses the wrong RFC number, so update every mention
of RFC 5822 to RFC 5322 in the mail guidance content. Make the change
consistently across the affected text in the mail affordance docs, especially
the sections that describe using existing EML source and creating a draft
without sending, so readers are pointed to the correct message-format spec.