Question 1

How to fix AI-DLC 'log review' error: "the reviewer appendix must contain exactly one Reviewer line matching the requested reviewer"?

Accepted Answer

This is a known bug in AI-DLC v2.11.0, fixed in v2.11.1 (via #2221). In 2.11.0 the reviewer-authority check in `core/tools/aidlc-lib.ts` counted every `**Reviewer:**` line, so when the review file began with the agent's identity marker (e.g. `**Reviewer:** aidlc-product-lead-agent`) plus the verdict block's own Reviewer line, it counted two and refused the review. Fix: upgrade AI-DLC to v2.11.1 or later (or update your pinned framework version). Since 2.11.1, a Verdict, Reviewer or Iteration line repeated word for word counts as one line, so a review file starting with the identity marker records normally with all its findings shown at the gate. A second Reviewer line naming a *different* agent is still correctly refused. Workaround if stuck on 2.11.0: instruct the reviewing agent (in the dispatch brief) to start the review file at `## Review` and include only one Reviewer line, inside the verdict block — not as the file's first line. Note: if you use `--retry-pending` after a refusal, the retry intentionally starts from a clean slot and does not carry the refused review's findings, so manually surface the first review's findings at the gate to avoid losing Major findings.

Question 2

How to fix "could not resolve the current routed unit (no diagnostic)" error when running aidlc engine state unit start?

Accepted Answer

This error occurs in aidlc 2.10.0 because the routing check inside `unit start` and `unit complete` spawns a child `orchestrate next` process and kills it after ~30 seconds. If `next` takes longer (common on large workspaces, e.g., 49,000+ files where it can run 20–125s), the start is refused with the misleading "no diagnostic" message.

**Fix:** Upgrade to aidlc 2.11.0 or later. PR #1369 raised the routing check timeout from ~30 seconds to 5 minutes, so a slow `next` (20–125s) no longer causes the refusal. This is the same fix that resolved issue #1547.

**Workarounds if you cannot upgrade immediately (Windows):**
1. Close heavy processes (Chrome, other Claude Code sessions, Visual Studio).
2. Temporarily remove the front-end `node_modules` folder.
3. Warm the file cache by running `aidlc engine orchestrate next` 2–3 times back to back.
4. Launch `unit start` via `Start-Process`, then every 300 ms for ~6 s set the marker process and its child `next` process PriorityClass to High.

Note: these workarounds only help if a warm `next` finishes under ~30s. If the refusal still appears on 2.11.0+, measure how long a plain `aidlc engine orchestrate next` takes and report it.

Question 3

Why do Kiro CLI subagents run on a different model than the one selected with /model?

Accepted Answer

Kiro CLI sub-agents (dispatched AI-DLC agents like composer, developer, reviewers) do not inherit the session model chosen with `/model` or `kiro-cli chat --model`. They always start on the saved default model (`chat.defaultModel` in `~/.kiro/settings/cli.json`), because Kiro sub-agents only inherit steering files, MCP servers, workspace access, spec state, and permissions — not the model. A `/model` switch only affects the current chat until you save it.

Fix: run `/model set-current-as-default` in Kiro CLI, or run `aidlc config models --session-model <model-id> --yes` before dispatching agents, so `chat.defaultModel` matches your intended model. This also prevents dispatch failures when the saved default model becomes unavailable ("model is unavailable, retrying now") while the conductor keeps working.

In AI-DLC v2.11.1 (PR #2405) the docs, `aidlc config models --show`, and `aidlc doctor` were corrected to state that agents start on the model saved in Kiro settings rather than claiming they inherit the session model.

Question 4

How to fix "could not acquire plugin build output lock" error when the plugin build hangs for 15 minutes?

Accepted Answer

The hang is a permissions problem, not lock contention. The shared lock folder `/tmp/aidlc-plugin-build-locks` is created with mode 0700 by whichever user builds first (e.g., root in CI), so any other user gets `EACCES`, which is silently retried every 25 ms until the 900000 ms (15 min) timeout. Workaround — give each user its own temp dir before running the build: ```sh mkdir -p "$HOME/.tmp" export TMPDIR="$HOME/.tmp" TMP="$HOME/.tmp" TEMP="$HOME/.tmp" bun /aidlc-plugin-build.ts ``` In CI where earlier steps run as root and later steps as an unprivileged user, set `TMPDIR`/`TMP`/`TEMP` to a folder inside the job's scratch directory for the unprivileged step. Alternatively, as the folder's owner, make the shared lock dir world-accessible with sticky bit: `chmod 1777 /tmp/aidlc-plugin-build-locks`. Verify with `ls -ld /tmp/aidlc-plugin-build-locks` — if it shows mode `drwx------` owned by another user, this is the cause.

Question 5

How to fix t242 test failure with "Refusing to complete feasibility" / SUMMARY_QUESTIONS_MISSING error?

Accepted Answer

The test `tests/unit/t242-state-transition-guard.test.ts` fails because its bare fixture has no stage-questions file, so the engine's `gate-start` refuses with `"reason_codes":["SUMMARY_QUESTIONS_MISSING"]` — the test is stale against a precondition added later, not broken logic. Two fixes:

1. **Recommended:** Set `AIDLC_SKIP_SUMMARY_CONFIRMATION_GUARD=1` in the test's spawned environment. This matches the precedent in `tests/unit/t-guard-refusal-actionable.test.ts` and keeps the test focused on the gate words. Verify with:
```bash
AIDLC_SKIP_SUMMARY_CONFIRMATION_GUARD=1 bun test tests/unit/t242-state-transition-guard.test.ts
# → 56 pass, 0 fail
```

2. **Alternative:** Seed a `feasibility-questions.md` file in the fixture so the real guard path is exercised — more faithful, but couples the test to the summary-confirmation flow.

Note: This failure is NOT caused by a leaking `CLAUDE_CODE_SESSION_ID` (that's the separate t265 issue); unsetting that variable will not fix it.

aidlc-workflows

Core Features

What It Can't Do

Use Cases

Tags

Getting Started

Similar Projects