Plan · CHG-2026-014914 · Agentic SDLC

change_id: CHG-2026-014914
risk_class: R3
autonomy_tier: A2
controls: ['TRC-01', 'CHG-04', 'HUM-14']
data_classification: internal
originator: olaf@freundcloud.com
agent_identity: svc-agent-platform
model_route: gateway/tier-frontier
supersedes: None

Approach

The reported symptom is one workflow step failing. The cause is a function two callers share, one of which is a person. Patching the workflow would leave hand allocation broken; patching the allocator fixes both, and is the smaller diff.

Write the test first, against the current code, so that the diagnosis is proven rather than assumed.

Files

scripts/draft_intent.py
scripts/tests/test_draft_intent.py
.github/workflows/06-operate.yml

Sequence

Write test_draft_intent.py asserting a reserved-but-unfiled id is skipped. → verify: it fails against the current allocator. If it passes, the diagnosis is wrong and everything below is premature.
Add reserved_change_ids() reading git refs; union it in main(). → verify: the test passes; next_change_id is still pure.
Move the push out of the fallback’s way in 06-operate.yml, and suffix a colliding branch name. → verify: actionlint/parse clean, and the issue body is honest about a failed push.
make build test lint gates. → verify: green, HUM-14 content with R3/A2.
Dispatch Stage 6 and watch it. → verify: completes, and the id it allocates is not one any existing branch holds.

Risks

Risk	Likelihood	Mitigation
The diagnosis is wrong again	medium — it already was once	Step 1 must fail before step 2. A test that passes first time disproves the theory
`git ls-remote` unavailable or slow in CI	medium	Failure returns what was found rather than raising; allocation degrades to today’s behaviour
The fix masks itself tonight	certain	Merging 014913 already moved main past the stuck branch. Verification asserts the mechanism, not just a green run
Declared R2 and refused by HUM-14	high if careless	Declared R3 for the workflow path floor; A2 is the matrix maximum at R3 in development
Suffixing the branch hides a real collision	low	It is a last resort behind R1, and the run number in the name makes it obvious in the branch list

Rejected

Reading the GitHub issues API for reserved ids. Needs a token, fails for a person working offline, and an id in an issue with no branch collides with nothing.
git push --force. It would go green tonight and destroy an untriaged finding.
Letting Actions open pull requests. It would make the fallback unnecessary by removing the control the fallback exists to respect.
Making the detector idempotent for an ongoing anomaly. A real problem, genuinely separate, and it would hide whether this fix worked.
Deleting the stranded stage6/* branches. Untriaged findings. Not a bug fix’s call.

Tests

scripts/tests/test_draft_intent.py — new
make build test lint gates, make negative
Stage 6 dispatched end to end

Rollback

Revert. The allocator returns to reading the working tree and the nightly returns to failing on a collision, which is the current state.