fix(cli): #2269 honor --dangerously-skip-permissions in hive-mind spawn

[JOhnsonKC201]

Fixes #2269.

Summary

hive-mind spawn --claude --non-interactive --dangerously-skip-permissions was silently dropping the flag. The arg parser (parser.ts normalizeKey) converts kebab-case to camelCase and stores only the normalized key, so flags['dangerously-skip-permissions'] was always undefined and the === true check always failed. The headless claude child therefore ran in permissionMode: \"default\", every Edit/Write landed in permission_denials, and the agent finished having changed nothing.

Fix

Mirror the dual-key fallback used a few lines above for isNonInteractive:

```ts
const skipPermissions =
(flags['dangerously-skip-permissions'] === true || flags.dangerouslySkipPermissions === true) &&
!(flags['no-auto-permissions'] || flags.noAutoPermissions);
```

This is the patch the issue reporter (@rvrheenen) proposed verbatim — kept narrow on purpose. Also resolves the documented default: true mismatch the secondary note in #2269 called out, because the parser populates the camelCase key with the option's default value.

Verification

Repro from the issue:

```
npx @claude-flow/cli hive-mind spawn --claude --non-interactive
--dangerously-skip-permissions -o "edit any file in this repo"
```

• Before: stream-json `init` event shows `"permissionMode":"default"`; every `Edit` lands in `permission_denials`; files unchanged.
• After: `--dangerously-skip-permissions` is appended to `claudeArgs`; spawned `claude` runs in `bypassPermissions`; edits succeed (matches running `claude -p --dangerously-skip-permissions` directly).

Scope

One file, one logical change. No behavior change for users who don't pass the flag (parser default behavior unchanged); only the previously-broken path is restored.

🤖 Generated with RuFlo

[ruvnet]

Thanks @JOhnsonKC201 — the kebab→camel diagnosis is right and the activation half of the fix works. Caught one regression while running the test locally before merging:

× skipPermissions=false when --no-auto-permissions is also passed
   → expected true to be false

Verified with the live parser:

new CommandParser({ allowUnknownFlags: true })
  .parse(['--dangerously-skip-permissions', '--no-auto-permissions']).flags
// →
{
  dangerouslySkipPermissions: true,
  autoPermissions: false,      // ← parser uses yargs-style negation
  ...
}

So neither flags['no-auto-permissions'] nor flags.noAutoPermissions is ever true; the parser stores the negation as autoPermissions: false. The predicate as written:

!(flags['no-auto-permissions'] || flags.noAutoPermissions)

…evaluates to !undefined = true, so the deny-half never fires. After your PR, --dangerously-skip-permissions --no-auto-permissions would skip permissions — strictly more permissive than the pre-fix state.

Suggested one-line fix:

const skipPermissions =
  (flags['dangerously-skip-permissions'] === true || flags.dangerouslySkipPermissions === true) &&
  !(flags['no-auto-permissions'] || flags.noAutoPermissions || flags.autoPermissions === false);

That keeps the dual-key fallback for the hand-built case and also covers the actual CLI form. The existing test on line 47 should then go green, plus please add one for the explicit camelCase: { dangerouslySkipPermissions: true, autoPermissions: false } → false.

Holding the merge until that lands. Everything else in the PR — the kebab→camel root-cause writeup, the dual-key idiom mirroring isNonInteractive, the parser-pinning test — is the right shape. Just the deny-half needs the third clause.

[JOhnsonKC201]

Thanks for catching the regression and pushing the fix. Good to know the parser stores --no-auto-permissions as autoPermissions: false rather than a camelCase key, and even better that the parser contract is pinned in tests now.