| name | slo-plan |
| description | Use this skill after /slo-architect (and /slo-tla if tla_required), when the user says "write the runbook", "plan the milestones", "turn this into tasks". Authors a full v4 runbook INTERACTIVELY, one milestone at a time, confirming each contract before moving on. Maximum 5 milestones per runbook — if scope needs more, suggest splitting. Refuses to generate the whole runbook in one shot; this is deliberate discipline, not a limitation.
|
/slo-plan — Write A V4 Runbook, Milestone By Milestone
Work one milestone at a time, confirming each contract block before the next.
Inputs
docs/slo/idea/<slug>.mddocs/slo/research/<slug>/synthesis.mdARCHITECTURE.md or docs/ARCHITECTURE.md; if missing, auto-generate a reality-first docs/ARCHITECTURE.md.- Optional
/slo-architect outputs: stack decision, interfaces, and if tla_required: true, verified design + TLC results.
Output
One file in the user's project: docs/RUNBOOK-<kebab-slug>.md. Use references/runbook-template_v_4_template.md first; the repo mirror is docs/slo/templates/runbook-template_v_4_template.md. The v3 template remains for old runbooks.
Discipline — The One Rule
NEVER generate a whole runbook in one shot. If asked, refuse: one-shot runbooks are syntactically valid and strategically thin. The interactive walk is the value.
Method
- Scaffold v4 metadata: ID, prefix, stack, tests, stable interfaces, red lines.
- Check architecture docs; generate a reality-first orientation doc if missing.
- Propose 2–5 milestones; split the runbook if scope needs more than 5.
- For each milestone, follow
references/methodology-milestone-authoring.md, then confirm scope, allow-list, and BDD specificity.
- Fill the Documentation Update Table, architecture diagram, and §5 Formal-Verification section. When the design has
kani_required: true, author the §5.8 Kani proof-obligation sub-block (target fn, property, bound, expected pre-fix/post-fix) and add a matching Kani-obligation row to each relevant milestone's Evidence Log; otherwise mark §5.8 N/A.
Contract Block Sentinels
Every milestone Contract Block includes the base rows plus:
- Exemplar code to copy: cite concrete paths or write
N/A — docs-only / N/A — no brownfield exemplar needed with a reason.
- Anti-exemplar code not to copy: cite risky legacy paths or patterns agents must not imitate; N/A with reason is valid.
- AI tolerance contract: required when the milestone introduces, modifies, or verifies AI/LLM behavior (
ai_component: true, prompt/tool-choice changes, eval harnesses, generated-output behavior, or AI-agent flows). Cite references/ai-tolerance-contract.md and fill accepted variance, deterministic boundary, eval evidence, retry / fallback, must-never outcomes, and bounded sample budget. For non-AI work, write N/A — no AI component.
- Data classification:
Public, Internal, Confidential, or Restricted; see references/proactive-controls-vocabulary.md.
- Proactive controls in play: cite stack-aware controls from
references/proactive-controls-vocabulary.md as actionable implementation constraints, using references/secure-construction-matrix.md to map each touched surface to secure defaults and tests.
- Abuse acceptance scenarios: cite
references/abuse-case-examples.md; required for every new surface. If no new surface, write N/A — no new surface introduced, see <reason>. Silent omission is forbidden.
- Measurement deliverables: required for any value-bearing milestone. Cite the runbook's §5A Measurement Contract and name this milestone's slice (events / runtime metrics / saved queries + guardrail owner + readout date). For non-value-bearing milestones, write
N/A — not value-bearing, see <reason>.
- Outcome Validation deliverables / Critical user journeys: required for any value-bearing milestone — cite §5C + the
cuj-<slug>-N ids; else N/A — not value-bearing.
Measurement Contract requirement
A runbook for a value-bearing feature MUST carry the §5A Measurement Contract before it is complete. "Value-bearing" is defined deterministically: the work introduces or changes user-facing capability. It EXCLUDES internal refactor, docs-only, and test-only work (those mark §5A N/A — not a value-bearing feature). Pull the per-feature inputs forward from the idea doc's ## Success thesis (/slo-ideate) and the /slo-product metrics feature measurement spec (feature_measurement_spec: true). If a value-bearing runbook reaches completion without a filled Measurement Contract, flag the gap and require it before handoff — but do not retroactively invalidate legacy runbooks authored before this section existed (they remain valid; the requirement is forward-looking, mirroring the §10 Carry-forward backward-compat posture).
Secure Value & Security Contract requirement
A value-bearing OR security-relevant milestone (security-relevant = touches identity, secrets, PII, payment, cloud, AI agents, public/network boundaries, CI/CD, or infrastructure) MUST carry the §5B Secure Value & Security Contract — populate Value Wedge, Operator Readiness, Threat Model Summary (cite the /slo-architect model), Security Test Plan (Bundles A–F), and the Detected Work Ledger (five dispositions → existing /slo-retro lanes, no new verb). Cite proactive controls by OWASP Proactive Controls 2024 name. The Operator Readiness Gate is enforced by /slo-execute from the M3 Secure Value Loop release onward. Forward-looking: flag a missing §5B; never invalidate legacy runbooks (mirrors §5A / §10). Non-security refactor/docs/tooling → N/A. Full authoring detail: references/secure-value-contract.md; canonical: docs/SECURE-VALUE-LOOP.md.
Outcome Validation Contract requirement. A value-bearing milestone (same deterministic definition as the Measurement Contract; EXCLUDES refactor / docs-only / test-only) MUST carry the §5C Outcome Validation Contract plus, in §17, its Outcome Scenarios (oc-<slug>-N), Critical User Journeys (cuj-<slug>-N), and a Core Capability Regression Matrix (every row pass | not_applicable | waived_with_reason, never blank). This is Outcome First Engineering: the promised user outcome is the primary Definition of Done — code completion alone is insufficient (§6.12). §5C Front-to-End Validation steps are per-layer (applicable | not_applicable(reason)) with ≥1 real cross-layer assertion — a single-layer / mock-only assertion does not satisfy it. Forward-looking like §5A / §5B: flag a missing §5C; never invalidate legacy runbooks. Full detail: references/outcome-validation-contract.md.
When the Refactor budget is anything except No refactor permitted beyond direct implementation, apply references/refactoring-discipline.md: behavior-preserving microsteps, pre-test evidence, and post-test proof.
BDD includes happy path, invalid input, empty state, dependency failure, retry/concurrency/persistence/backward compat as applicable, and abuse case for new surfaces. For value-bearing milestones, Outcome Scenarios are outcome-shaped (one observable user outcome + follow-on Ands) and are the primary Definition of Done — they are distinct from, and rank above, the per-scenario BDD Acceptance table.
Gates
Refuse when file ownership is unclear, BDD is generic, Definition of Done or Evidence Log is absent, or Forbidden shortcuts is empty. For a value-bearing milestone, also refuse when: an Outcome Scenario is vacuous / single-And / mock-only (theatre), the §5C Front-to-End path is monolithic instead of per-layer (no not_applicable(reason) granularity, no cross-layer assertion), or a Core Capability Regression Matrix row is blank (every row must be pass | not_applicable | waived_with_reason).
Anti-Patterns
- Generic BDD, 30-row allow-lists, deferred Evidence Logs, scope beyond 5 milestones, silent omission of the three security rows, vague abuse cases, or new data/control vocabulary.
Handoff
After the runbook is complete, suggest /slo-critique before execution starts.
Loops: Sprint loop — see docs/LOOPS-ENGINEERING.md#sprint-loop.
