STANDARD OPERATING PROCEDURE
Purpose
Create and supervise child tasks without uncontrolled fan-out, keeping registry, quotas, and evidence tracking aligned with confidence ceilings.
Trigger Conditions
- Positive: spawning sub-tasks, parallelizing work, delegating to specialized agents, burst control, and watchdog setup.
- Negative: single-thread execution, prompt-only edits (route to prompt-architect), or new skill creation (route to skill-forge).
Guardrails
- Skill-Forge structure-first: maintain
SKILL.md, examples/, tests/; add resources//references/ or document remediation tasks.
- Prompt-Architect hygiene: record intent, HARD/SOFT/INFERRED constraints, and provide pure-English outputs with ceilings.
- Spawn safety: cap concurrency, enforce registry-only agents, define quotas/timeouts, and set rollback/abort paths; honor hook latency budgets.
- Adversarial validation: test runaway prevention, failure retries, and cancellation flow; capture evidence.
- MCP tagging: persist spawn runs with WHO=
safe-task-spawn-{session} and WHY=skill-execution.
Execution Playbook
- Intent & limits: capture goal, constraints, quotas, and safety thresholds; confirm inferred limits.
- Plan & registry: define child tasks, assign owners, and register agents with namespaces.
- Spawn & supervise: launch tasks with timeouts/backoff; track TodoWrite and health.
- Safety nets: set circuit breakers, cancellation rules, and rollback steps.
- Validation loop: simulate overload, timeout, and failure scenarios; log telemetry and evidence.
- Delivery: report tasks, outcomes, risks, and confidence ceiling.
Output Format
- Objective and constraints with quotas/timeouts.
- Task list with owners, schedules, and safety rules.
- Validation evidence (overload, cancel, retry) and residual risks.
- Confidence:
X.XX (ceiling: TYPE Y.YY) - rationale.
Validation Checklist
- Structure-first assets present or ticketed; examples/tests aligned to spawn controls.
- Quotas, timeouts, and registry enforced; rollback/abort paths validated; hooks within budget.
- Adversarial/COV runs stored with MCP tags; confidence ceiling declared; English-only output.
Completion Definition
Spawning is complete when tasks finish or are cleanly aborted, evidence is stored, quotas respected, and risks owned with next steps logged.
Confidence: 0.70 (ceiling: inference 0.70) - Safe spawning SOP rebuilt with skill-forge structure and prompt-architect constraint and confidence discipline.
1---2name: safe-task-spawn3description: Spawn and supervise tasks safely with registry controls, resource quotas, and rollback-ready delegation.4---5
6
7
8
9## STANDARD OPERATING PROCEDURE
10
11### Purpose
12Create and supervise child tasks without uncontrolled fan-out, keeping registry, quotas, and evidence tracking aligned with confidence ceilings.
13
14### Trigger Conditions
15- **Positive:** spawning sub-tasks, parallelizing work, delegating to specialized agents, burst control, and watchdog setup.
16- **Negative:** single-thread execution, prompt-only edits (route to prompt-architect), or new skill creation (route to skill-forge).
17
18### Guardrails
19- **Skill-Forge structure-first:** maintain `SKILL.md`, `examples/`, `tests/`; add `resources/`/`references/` or document remediation tasks.
20- **Prompt-Architect hygiene:** record intent, HARD/SOFT/INFERRED constraints, and provide pure-English outputs with ceilings.
21- **Spawn safety:** cap concurrency, enforce registry-only agents, define quotas/timeouts, and set rollback/abort paths; honor hook latency budgets.
22- **Adversarial validation:** test runaway prevention, failure retries, and cancellation flow; capture evidence.
23- **MCP tagging:** persist spawn runs with WHO=`safe-task-spawn-{session}` and WHY=`skill-execution`.
24
25### Execution Playbook
261. **Intent & limits:** capture goal, constraints, quotas, and safety thresholds; confirm inferred limits.
272. **Plan & registry:** define child tasks, assign owners, and register agents with namespaces.
283. **Spawn & supervise:** launch tasks with timeouts/backoff; track TodoWrite and health.
294. **Safety nets:** set circuit breakers, cancellation rules, and rollback steps.
305. **Validation loop:** simulate overload, timeout, and failure scenarios; log telemetry and evidence.
316. **Delivery:** report tasks, outcomes, risks, and confidence ceiling.
32
33### Output Format
34- Objective and constraints with quotas/timeouts.
35- Task list with owners, schedules, and safety rules.
36- Validation evidence (overload, cancel, retry) and residual risks.
37- **Confidence:** `X.XX (ceiling: TYPE Y.YY) - rationale`.
38
39### Validation Checklist
40- Structure-first assets present or ticketed; examples/tests aligned to spawn controls.
41- Quotas, timeouts, and registry enforced; rollback/abort paths validated; hooks within budget.
42- Adversarial/COV runs stored with MCP tags; confidence ceiling declared; English-only output.
43
44### Completion Definition
45Spawning is complete when tasks finish or are cleanly aborted, evidence is stored, quotas respected, and risks owned with next steps logged.
46
47Confidence: 0.70 (ceiling: inference 0.70) - Safe spawning SOP rebuilt with skill-forge structure and prompt-architect constraint and confidence discipline.
majiayu000