Follow-up: deferred review findings from PR #845

[philcunliffe]

Triage of the residual review findings on PR #845 (head 590d4f438466daeca01285d5d270a8fe13814903, branch fix/issue-835) classified every open finding as non-blocking, so the PR can merge and these are deferred here. Each item carries its evidence from the review rounds, re-verified against the head in a clean worktree.

Deferred findings

1. hyp help <group> <typo> renders the group help and exits 0 (round 2, finding 1, low)

src/core/cli/dispatch.js:509: the helpVerb branch renders group help and sets exitCode = 0 regardless of leftover tokens in matched.rest, so hyp help query bogus prints the query group table and exits 0, while hyp query bogus --help exits 2 (src/core/cli/group_help.js:157) and hyp help graph bogus exits 2. Same user error, three answers.

Why deferred: the page shown is the right command's page and lists the valid subcommands, so the typo is self-correcting; nothing wrong executes. The obvious guard (refuse leftover non-flag tokens) is wrong on this codebase because backfill is a group parent that legitimately takes positionals (src/core/cli/core_commands.js:150-153), and letting the group command run instead would reintroduce the hyp help purge <path> hazard (src/core/commands/backfill.js:50 does not handle a stray --help). A correct fix needs a way to tell a pure group (built by makeGroupCommand) apart from a group whose bare command takes positionals: a new marker on CommandRegistration or a WeakSet exported from src/core/cli/group_help.js. That is new public API surface plus a design decision LLP 0270 does not settle.

2. hyp help zzz reports unknown command 'help zzz' (round 2, finding 2, nit)

src/core/cli/dispatch.js:404 echoes typedArgv.join(' '), so the diagnostic quotes the pair as though a help namespace exists, while the token that is actually unknown is zzz. The current wording is deliberate (it echoes exactly what the user typed) and is asserted by test/core/cli/help-metadata.test.js:195. If the other spelling is preferred, typedArgv.slice(1) on this path is the whole change, plus the test update.

3. PR body understates the test file and misreports the suite baseline (round 2, finding 3, nit)

The PR body says 11 assertions with 10 of 11 failing on master; test/core/cli/help-metadata.test.js contains 14 test(...) cases. The body also reports 4207 pass / 22 fail as the local baseline, while a clean-environment run on the same head gives 4497 pass / 0 fail, so the 22 failures were environmental. No code impact; a body edit or a note in the merge commit closes it.

4. Blanket --platform refusal also rejects the matching host (round 1, finding 6, design call)

src/core/commands/daemon.js:457: the rule is p.platform !== undefined && !p['dry-run'], so hyp daemon install --platform linux is refused on Linux, where it was a correct no-op-equivalent. LLP 0270 #platform-is-a-render-override rests its rationale on cross-platform breakage only; a rule of p.platform !== process.platform would be exactly as safe. The blanket form is defensible (mirrors --json requires --dry-run) and is settled in an Accepted LLP, so changing it means a new LLP extending 0270, either narrowing the rule or restating the rationale as "any --platform, matching or not". Exit is a clean 2 with a clear message and no in-repo caller is broken, so this is a documentation-honesty and ergonomics call, not a defect.

5. runDaemonInstall still prints its own usage string (round 1, finding 7, informational)

src/core/commands/daemon.js:178 keeps a hand-maintained usage: line reachable when --help is not the first token after the command (hyp daemon install --dry-run --help). Pre-existing tension with LLP 0009 "a command body never prints its own usage"; the fix is a cross-command cleanup in parseCommandArgv's help branch, out of scope for #835.

Not carried forward

Round 1 finding 8 (branch behind origin/master) is resolved: head 590d4f4-era merge brought the branch to 0 commits behind at triage time.

Backlink: PR #845 (fixes #835). Review records: the two neutral-review comments on that PR, for heads 649f02b3 and 590d4f43.

[philcunliffe]

Blocked on PR #845, which is still open and unmerged. Four of the five deferred findings point at code that does not exist on master, so there is nothing on master to reproduce against and no honest base for a fix branch.

What neutral did

Read the issue, then checked every referenced path against origin/master (b1afa7c) before touching a worktree. PR #845 (fix/issue-835, head 590d4f4) is OPEN, mergedAt: null, and its diff touches exactly the files these findings cite:

README.md
llp/0009-cli-registry.spec.md
llp/0270-help-metadata-is-part-of-the-contract.decision.md
src/core/cli/core_commands.js
src/core/cli/dispatch.js
src/core/commands/clients.js
src/core/commands/daemon.js
test/core/cli/help-metadata.test.js

Finding-by-finding ground truth against master (b1afa7c)

#	Finding	Status on master
1	hyp help <group> <typo> renders group help, exits 0 (dispatch.js:509)	Not reproducible. There is no helpVerb branch on master. grep -n 'helpVerb' src/core/cli/dispatch.js on origin/master returns nothing; help appears only as a member of HELP_FLAGS (dispatch.js:42). The whole hyp help <group> rendering path arrives with #845.
2	hyp help zzz quotes the pair (dispatch.js:404, typedArgv.join(' '))	Not reproducible. No typedArgv symbol exists anywhere in master's dispatch.js; the diagnostic there is dispatch.js:387, hyp: unknown command '${argv.join(' ')}'. The asserting test, test/core/cli/help-metadata.test.js, is not in master's tree at all (test/core/cli/ contains only init-help-flags, pick-gate-disclosure, spinner, stream-errors, style, and the tui/ files).
3	PR body understates the test file and misreports the baseline	Not a code change. The artifact to correct is PR #845's own body, on an open PR that is still in review.
4	Blanket --platform refusal also rejects the matching host (daemon.js:457)	Not reproducible. master's parseDaemonInstallArgv (src/core/commands/daemon.js:433-450) has no refusal rule at all: it validates platform against the ['darwin','linux'] enum and returns it. There is no p.platform !== undefined && !p['dry-run'] guard to narrow. The issue also scopes the fix as "a new LLP extending 0270" and llp/0270-* does not exist on master either.
5	runDaemonInstall prints its own usage string	Present on master (src/core/commands/daemon.js:258, plus siblings at :338 and :407), but the issue itself classifies it as informational, pre-existing, and "a cross-command cleanup in parseCommandArgv's help branch, out of scope for #835". It is not the subject of this follow-up.

Why neutral stopped rather than shipping something

The ground-truth gate is a test that fails on the current code and then passes. For findings 1, 2, and 4 no such test can be written against master: the branches, symbols, and refusal rule under discussion are introduced by #845. Basing a branch on origin/fix/issue-835 instead would stack a speculative PR on an unmerged head that may still change under review, and finding 4 in particular depends on an LLP (0270) that is not yet part of the corpus, so any extending LLP would reference a document that does not exist on the default branch.

What neutral needs from a human

1. Merge PR Core CLI help metadata is stale and incomplete (#835) #845 first (preferred). Once Core CLI help metadata is stale and incomplete (#835) #845 lands on master, findings 1, 2, and 4 become reproducible and neutral can take this issue end to end: reproduce, fix, PR. Re-apply neutral:fix (or just comment) after the merge and neutral will pick it up on its next tick.
2. Or authorize stacking. If Core CLI help metadata is stale and incomplete (#835) #845 is going to sit open for a while and you want these fixed in parallel, say so and neutral will branch from origin/fix/issue-835 and open a PR targeting fix/issue-835 rather than master, accepting the rebase risk.
3. Or scope this issue down. If you want only finding 5 (the one thing that does exist on master) addressed now, say so - though note the issue text explicitly rules it out of scope, and it is a cross-command cleanup touching parseCommandArgv, not a Core CLI help metadata is stale and incomplete (#835) #845 follow-up.
4. Finding 3 needs a decision regardless of the above. It is a PR-body correction on Core CLI help metadata is stale and incomplete (#835) #845 (14 test(...) cases, not 11; clean-environment baseline 4497 pass / 0 fail, not 4207 / 22). neutral did not edit the body of an open PR under review unilaterally. Tell it to, and it will - or fold the correction into the merge commit message.

Also worth noting for whoever picks this up: the issue's line numbers are anchored to #845's head 590d4f4, not to master. Any future re-triage should re-resolve them against whatever head is current then.

Reply with a comment on this issue and neutral will re-engage with your guidance on its next tick.