Restore /look:again and /look:tidy under /look: namespace

[pike00]

Summary

• Migrate /look:again and /look:tidy back from src/skills/ to src/commands/ so they appear under the /look: namespace in the slash-command picker (matching the README) instead of as bare /again and /tidy.
• The underlying bug that drove the 0.4.0 commands/ -> skills/ migration (multi-command plugins misrouting on duplicate name: frontmatter) was fixed upstream in Claude Code 2.1.101: "Fixed several plugin issues: slash commands resolving to the wrong plugin with duplicate name: frontmatter ..." The skills workaround is no longer needed.
• Defense-in-depth: the name: frontmatter field is now omitted from both command files, so the slash name derives from the filename. This removes the duplicate-name collision vector entirely, regardless of resolver version.
• lookagain-output-format remains a skill - the reviewer subagent invokes it via the Skill tool; moving it to commands/ would break the again workflow.
• Version bumped 0.5.0 -> 0.5.1.

Migration history

• 1d15144 (0.4.0, 2026-01-29) - workaround migration, commands/ -> skills/.
• This PR (0.5.1) - revert the workaround now that upstream is fixed.

Test results

Structural (make test)

=== Results: 59 passed, 0 failed ===

Behavioral (make eval via promptfoo)

14 / 14 passed against anthropic:messages:claude-sonnet-4-6. Total cost $0.6067, duration 1m 43s, 58,156 tokens (14,680 prompt + 23,895 completion + 19,581 cached).

#	Test	Result	Latency	Cost	Rubric finding (truncated)
1	no arguments -> defaults	pass	40.0s	$0.0501	Correctly applies passes=3, target=staged, auto-fix=true, model=thorough, max-passes=7. Describes applying must_fix fixes; does not complain about missing arguments.
2	auto-fix=true -> applies fixes	pass	44.1s	$0.0541	Output repeatedly states it will apply fixes for must_fix issues between passes.
3	auto-fix=false -> skips fixes	pass	41.5s	$0.0499	Output repeatedly states no fixes will be applied.
4	passes=5 -> 5 initial passes	pass	48.6s	$0.0489	Assertion passed.
5	model=fast -> haiku	pass	40.6s	$0.0524	Assertion passed.
6	model=thorough -> no override	pass	cached	$0.0541	Output explicitly omits the model parameter from the Task tool call.
7	target=branch -> branch scope	pass	43.7s	$0.0548	Output uses "scope: branch - review all changes on this branch vs base".
8	only auto-fix=false -> rest default	pass	41.5s	$0.0489	Correctly parses auto-fix=false and applies all other defaults; no fixes applied.
9	summary tables include Suggested Fix + global ID	pass	cached	$0.0501	Assertion passed.
10	aggregate.md preserves full Suggested Fix; conversation may truncate	pass	cached	$0.0501	Output explicitly states "full text is always in aggregate.md and aggregate.json".
11	must_fix table has Fixed column; should_fix/suggestions do not	pass	cached	$0.0501	Confirms must_fix has the Fixed column and the other two tables omit it.
12	tidy all=true -> removes all runs	pass	14.5s	$0.0155	Output skips any date/cutoff calculation when all=true.
13	tidy keep=3 -> date-based retention	pass	13.9s	$0.0148	Calculates cutoff = today - 3 days, removes only directories older than cutoff.
14	tidy no arguments -> keep=1, all=false	pass	11.5s	$0.0131	Uses defaults; describes 1-day retention with date-based filtering.

Full result JSON is 417 KB (not attached). To regenerate locally:

nvm use 22.22.0   # or any node >=22.22.0 / ^20.20.0
make eval

[HartBrook]

Looks good!