`) or work + # If this fails: either re-record (`npm run record -- --all --agents [--board-writes] --provider
`) or work # out why the prompt moved. Do not silence it. - name: no cassette drift run: | set -uo pipefail drift=0 - for mode in "" "--twice" "--provider anthropic" "--agents" "--agents --provider anthropic"; do + for mode in "" "--twice" "--provider anthropic" "--agents" "--agents --provider anthropic" \ + "--agents --board-writes" "--agents --board-writes --provider anthropic"; do # shellcheck disable=SC2086 n=$(npm run demo --silent -- $mode 2>&1 | grep -c "recorded against a different prompt" || true) echo "demo ${mode:-(default)}: $n drifted cassette(s)" From 5580ec4cd6e23f9c98077049918f5ca5a1701c37 Mon Sep 17 00:00:00 2001 From: harishghasolia07 <100846446+harishghasolia07@users.noreply.github.com> Date: Wed, 26 Aug 2026 15:43:59 +0000 Subject: [PATCH 2/2] Say plainly that no board-write recording shows a gate refusing MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit I described this mode as the one that demonstrates an agent being refused by a gate. Checked it: across all eight scenarios on both providers, --board-writes produces zero refused writes. The claim was wrong. The first recording did show five refusals, which is where the impression came from — but they were all `unresolvable field(s): FINAL_DESC`, and that was the tool spec's fault, not the gate catching a bad model. create_task listed `description` as optional while the gate requires it. Fixing the schema took refusals to zero, which is the correct outcome and also removes the only footage of a refusal. So the same disclosure the role-agent recordings already carry now applies here: the path is wired and exercised end to end, no recording shows it firing, and what proves it is governedTracker.test.ts with scripted replies — off-roster assignee, unknown list key, credential-touching title, uncited subtask, and an op with no manifest form, each refused with the inner adapter asserted never to have seen it. A green --board-writes run is evidence the mechanism runs, not evidence it bites. --- AGENTS.md | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/AGENTS.md b/AGENTS.md index b81ecd4..e726172 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -136,6 +136,21 @@ happened to say on one day; if the feature needed a model to disagree in order t the demonstration would be the weather. But it does mean the honest claim is **"the path is proven by test, not by recording"** — and a reader who wants to see it fire should run the tests, not the demo. +**The same is true of the board-write recordings, and for a reason worth recording.** Across all eight +scenarios on both providers, `--board-writes` produces **zero refused writes**. Every write the board +agent originates passes the gates. + +The first recorded attempt was not like that: it produced five refusals in a single scenario, all of +them `unresolvable field(s): FINAL_DESC`. That was not the model failing the gate — `create_task` +listed `description` as optional while the gate requires it, so the tool was lying about what a valid +write looks like and the model believed it. Fixing the schema took the refusals to zero. + +Which leaves the same honest position as above: the governed write path is wired and exercised end to +end, and **no shipped recording shows a gate refusing a write.** What proves it does is +`governedTracker.test.ts` with scripted replies — an off-roster assignee, an unknown list key, a +credential-touching title, a subtask whose parent history was never read, and an operation with no +manifest form at all. Each is refused, and the inner adapter is asserted never to have seen it. + ### This is what "authority to write" means The internal spec this repo was built from describes the Board agent as *"the orchestrator above the