[diillson]

What

In coder mode, a response whose tool call came out as <invoke name=...>/<parameter name=...> markup (the function-call format some models fall back to), or as a run of stray </invoke>, carried no call chatcli recognizes. The loop took the text for the final answer and printed "Task completed" with the action never run.

Now the response is checked for that markup outside code fences and inline code. If it is there, the model is told that nothing was executed, shown the <tool_call> format, and asked to resend, at most three times in a row (the counter resets when a turn acts).

The markup is deliberately not converted into a call. Text that only looks like a call (quoted from a file or a tool result) must never run because it appeared in a response; the correction costs one turn and leaves every policy gate untouched.

Why

Found by reading the failures of a LongMemEval-V2 run: 33 of 484 coder attempts (7%) ended without the file the agent had been asked to write, and 29 of those 33 ended with exactly this markup as the final message. Among the 451 attempts that finished, 1 did.

Evidence

• Offline, over the 484 saved final messages: the detector flags 29 of the 33 failures (the other 4 have no markup at all, a different cause) and 1 of the 451 successes. That one had already written its file, so it would cost one extra correction turn.
• End to end with the real binary against a scripted fake model that leaks the markup first: the previous binary prints "Task completed" and the file does not exist; this one corrects, the model resends, the command runs and the file contains the expected text.
• Unit tests cover the four shapes seen, the cases that must not fire (markup inside a fence or inline code, a valid tool call, prose, a single stray closer) and that the correction never echoes the markup. golangci-lint v2: 0 issues; go test ./cli/... passes.

[github-actions]

Quality Gate

Result: ✅ all floors passed

Floor	Status	Result	Δ vs main	Budget
1 · Build & Static	✅	go build / vet / fmt / lint	—	—
2 · Coverage	✅	69.6% (bootstrap)	0	≥ baseline
3 · Patch coverage	⚠️	67.9% (req ≥ 60%)	—	≥ 60%
4 · AI smells	✅	diff scanned	—	—
5 · Scope budget	✅	3 files / 162 LOC (code 162 + tooling 0)	—	warn 800·25
6 · E2E	✅	go test -race ./e2e/...	—	≤ 15min
7 · Commit lint	✅	conventional commits	—	—
8 · Cyclo (new code)	✅	1 file(s) under threshold	—	≤ 30
9 · Secrets scan	✅	gitleaks	—	—
10 · i18n parity	✅	missing 0, unknown 0	—	—
11 · CRD drift	✅	drifted: 0	—	—
12 · License headers	✅	0 missing	—	—
13 · API breaking	✅	0 incompatible	—	—
14 · Binary size	✅	chatcli 100.4MB · operator 55.4MB	—	100MB each
15 · Provider parity	✅	14 providers · 0 violations	—	—

Config: .github/quality-gate.yml. Workflow: .github/workflows/quality-gate.yml.