Three reproducible bugs reported by @grym3s, batched in the style of #2346.
#2352 — hooks post-edit: Windows paths rejected, failure printed as [OK]
- validatePath used the general SHELL_META set which includes `\`, so
every absolute Windows path (E:\Repos\...) failed with "shell
metacharacters". Claude Code hook events deliver absolute paths in
tool_input.file_path, so all Windows users had silent learning loss.
Path-specific PATH_SHELL_META drops `\` (still rejects ; & | ` $ etc.)
and the sanitized output normalizes to forward slashes.
- CLI post-edit action printed `Outcome recorded for ...` whenever the
MCP call returned at all, ignoring `{success:false}`. Now checks
result.success, surfaces the error, and exits non-zero.
#2351 — trajectory-end: step-less feedback never distilled
- When `trajectory-end` is called with `feedback` but no recorded steps,
the feedback was persisted with the trajectory but never embedded as a
searchable pattern — patternsExtracted always reported 0 and
pattern-search never surfaced it. Step-less trajectories are the
common case for LLM agents. Now routes feedback through the same
bridge.bridgeStorePattern / store-fallback path that pattern-store
uses, with modest default confidence (0.6 success / 0.4 failure)
since step-less feedback hasn't been validated by execution evidence.
Best-effort: distillation failures do not fail the trajectory-end call.
#2350 — init hooks: subcommand writes no hooks block
- settings-generator (#1744 fix) only emits the `hooks` block when
`components.helpers` is true, because hooks point at the helper
handler script. The `init hooks` subcommand was setting helpers:false,
so the one subcommand whose purpose is "Initialize only hooks
configuration" produced settings.json with no `hooks` key while
reporting "N hooks enabled". Helpers must ship with hooks.
Tests
- v3/@claude-flow/cli/__tests__/validate-input-path-2352.test.ts (22 tests):
pins Windows-path acceptance, POSIX-path acceptance, and rejection of
shell metacharacters, traversal, empty/non-string/oversized input.
- All existing validate-input, init-wizard-bugs, hooks-intelligence, and
hooks-post-task tests still pass.
rUv