Commit d3d0b6a
fix(qodo): address PR #42 review findings
Addresses Qodo automated review round 1 on PR #42.
## Finding #2 — stub continuation exception ambiguous (ADDRESSED)
`system-prompts/orchestrator_stub.md`:
- Narrow the exception trigger to the deterministic "PLAN ALREADY
APPROVED" / "continuing to STAGE 1" continuation marker only.
Dropped the "OR most recent tool call was ExitPlanMode" clause
because it could spuriously match a fresh user prompt arriving
right after a prior plan approval.
- Explicitly whitelist the tools permitted in the exception path:
`Agent`, `TaskCreate`/`TaskUpdate`/`TaskGet`, and `TeamCreate`/
`SendMessage` (for team mode). This removes ambiguity with the
stub's `## Rule` section that otherwise restricts tools to
Tasks API / AskUserQuestion / delegate.
## Finding #1 — continuation sentinel removed (INTENTIONALLY UNCHANGED)
`hooks/stop/python_stop_hook.py` is left as-is. Per the Claude Code
Stop hook docs, `reason` is the only field injected into the model's
continuation context; `systemMessage` is user-facing only. Moving the
anti-redelegation directive into `systemMessage` (as suggested) would
make it invisible to the model and defeat Layer 1 of the loop fix.
Repo-wide grep also confirms zero downstream readers of `reason` — the
literal "continue" was label text, never a parsed sentinel. Full
defense-of-design posted as a PR comment on #42.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>1 parent 32d9eaa commit d3d0b6a
1 file changed
Lines changed: 1 addition & 1 deletion
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
12 | 12 | | |
13 | 13 | | |
14 | 14 | | |
15 | | - | |
| 15 | + | |
16 | 16 | | |
17 | 17 | | |
18 | 18 | | |
| |||
0 commit comments