Stop and escalate
Stop and Escalate
Section titled “Stop and Escalate”Agent instruction protocol: 1.0.
Behavioral compliance depends on the connected agent or integration. The Day Shift CLI exposes repository policy and evidence, but it cannot force an unsupported agent to obey them.
Stopping is a normal lifecycle result when the next safe action needs evidence or authority the current request does not supply.
Block advancement when
Section titled “Block advancement when”- selected work or its current revision is ambiguous;
- readiness, required dependency, baseline, review, receipt, or validation evidence is missing, stale, partial, or contradictory;
- the requested change expands beyond declared scope or protected paths;
- implementation would overwrite unrelated user changes;
- required validation fails or cannot meaningfully evaluate acceptance;
- a destructive action, release, publication, credential, network, or external-system effect needs separate authority;
- a review result is being treated as mutation, disposition, acceptance, or closeout permission;
- discovered work would require a new artifact, sibling, milestone, phase, or priority decision;
- a running command has not reached a terminal result.
A blocked acceptance gate does not automatically require human permission or end execution. For an authorized task implementation, classify continuation using .day-shift/workflows/task.agent-lifecycle-guide.md#recovery-decisions: wait on the same running handle, use a bounded exact retry, or repair through the eligible owning task. Callable guidance never grants authority. Escalate only when no authorized eligible recovery remains, and preserve stricter user restrictions.
Do not reorder required siblings, hide the blocker in a summary, fabricate evidence, broaden scope to keep momentum, or retry a still-running or already-completed command merely to obtain cleaner output.
Escalation record
Section titled “Escalation record”Report:
- Selected work: exact repository-relative artifact and relevant revision.
- Blocked action: the transition or outcome that cannot safely continue.
- Evidence: the concrete finding, missing dependency, failed command, conflict, or authority boundary.
- Preserved state: changed paths, live process handle, transaction or receipt identity, and whether any write occurred.
- Narrowest resolution: the exact decision, artifact repair, authorization, or external-state change needed.
- Safe alternatives: only alternatives that remain inside current scope and authority.
Ask for a decision only when repository evidence cannot resolve it. If the task, workflow, or current user instruction already establishes the answer, state that contract and continue.
Discovered work
Section titled “Discovered work”Record out-of-scope discoveries in the current task evidence or reconciliation under the declared taxonomy. Reporting discovered or follow-up work is read-only; creating planning artifacts requires a separate authorized planning action.
Resume
Section titled “Resume”After the blocker is resolved, revalidate the selected artifact and existing run through Current work and resume at the last fresh gate. Repeat Agent orientation only when the execution environment or selection requires it. Use Review boundaries before consuming any saved finding or receipt.