Validating

Validating

If preflight readiness is missing, stale, or blocked (check .pulse/tooling-status.json), stop and invoke pulse:using-pulse.

Purpose

Do not start execution unless current work is feasibility-ready in real repo conditions. This skill verifies that:

• mode and shape still fit current reality
• assumptions are proven with concrete evidence
• current-work artifacts are executable
• beads are schema-valid and implementation-ready when needed
• execution stays blocked until explicit user approval

Non-Negotiable Gates

• Gate A — Approved shape is required in work-shape.md, phase-plan.md, or epic-map.md.
• Gate B — Reality + feasibility must pass before structural polish.
• Gate C — If beads exist, bead schema must pass before structural checker.
• Gate D (Gate 3) — Explicit user approval is mandatory before execution.

If any gate fails, stop and route back; never “proceed anyway.”

Required Inputs

• history/<feature>/CONTEXT.md
• history/<feature>/discovery.md
• history/<feature>/approach.md
• approved shape artifact: work-shape.md, phase-plan.md, or epic-map.md
• mode-required current-work artifacts:
  • direct/spike/small: current work in work-shape.md
  • epic-map: current-story-pack.md
  • phase-plan: phase-<n>-contract.md and phase-<n>-story-map.md
• .beads/*.md for current work only when beads already exist

If any required non-bead artifact is missing, return to pulse:planning.

Runtime Flow (Hot Path)

0) Orient to mode, shape, and current work (mandatory every run)

Read both:

• approved shape artifact (source of truth)
• .pulse/STATE.md (mirror)

Confirm and present:

• mode
• shape artifact and approval status
• current work and practical goal
• mirror status (in sync / out of sync / missing)

If shape approval is not APPROVED, stop. If shape artifact and mirror disagree, stop and return to planning for sync.

Use orientation template from references/runtime-appendix.md.

1) Reality gate (fail fast)

Run mode-fit and repo-fit checks before structural validation:

• mode still matches task reality
• approved shape matches real constraints
• assumptions are explicit and testable
• no smaller safe path is being overruled without reason
• proof surface exists (files/APIs/commands/runtime checks)

If reality gate fails, route back to planning.

2) Feasibility matrix + spikes/probes

Build feasibility matrix for blocking assumptions:

• always for high_risk_feature
• for standard_feature whenever meaningful assumptions remain

Require spikes for unproven assumptions that can invalidate current work:

• one spike bead = one yes/no spike_question
• isolated execution with 30-minute timebox
• findings at .spikes/<feature>/<spike-id>/FINDINGS.md
• definitive YES or NO

If timebox expires without decision:

• present findings
• offer +15m extension only with explicit approval, replan, or mitigation
• never continue silently

Decision handling:

• YES: require planning-owned artifacts/beads to reflect constraints before execution approval (route to planning when those artifacts must change)
• NO: stop, update planning artifacts, return to planning, then re-run validating

Bead-absence branch (feasibility-first path):

• If feasibility is READY or READY WITH CONSTRAINTS and beads are required for current work but do not yet exist, stop and route to pulse:planning to create only validated current-work beads.
• After bead creation, resume pulse:validating at Schema Gate, then continue structural checks and bead review before Gate 3 approval.

3) Schema gate (before structural checker)

Run schema gate only when current-work beads exist.

Before any structural checker:

• verify every bead has canonical required fields
• verify testing_mode: tdd-required beads include red/green tdd_steps
• verify verify and verification_evidence are concrete
• verify file scope is not obviously overbroad
• verify HIGH-risk beads include meaningful learning_refs when prior recall clearly applies

If schema fails, repair bead schema defects when local/clear; route to planning when the fix requires reshaping planning-owned artifacts.

Use schema checklist in references/runtime-appendix.md.

4) Structural verification (max 3 iterations)

Run plan-checker only after feasibility + schema pass. PASS only if all dimensions pass:

1. mode/shape coherence
2. current-work coverage and ordering
3. decision coverage
4. dependency correctness
5. file scope isolation
6. context budget
7. verification completeness
8. integration/exit-state/risk coherence

Rules:

• PASS only if all dimensions pass.
• On fail, fix exact artifacts and re-run.
• Maximum 3 iterations. If still failing after 3, escalate and stop.

5) Bead polishing and readiness

When beads are required, polish bead graph:

• bv --robot-suggest for dependency completeness
• bv --robot-insights for cycles/bottlenecks/orphans
• bv --robot-priority for foundational ordering
• deduplicate overlapping same-goal same-scope beads
• run fresh-eyes bead review and resolve all CRITICAL flags
• confirm current-work ↔ bead mapping coherence both directions

Then run readiness check:

• entry state is concrete
• exit state is observably provable
• integration path is credible
• verification is concrete
• architecture boundaries are preserved
• scopes remain surgical

Any “no/not sure” routes back to planning artifacts or bead set; do not approve.

6) Final approval gate (Gate 3 hard stop)

Present validation summary and request explicit approval. Use structured question tools when available; otherwise plain text.

Execution is forbidden until approval is explicit.

On approval:

• update .pulse/STATE.md and .pulse/state.json to validated status
• default approved path: record gate: GATE 3, gate_status: approved, next_action: manual_invoke, and set next_skill_recommended from recommended_mode, then stop
• optional fast path: only continue in the same session when the user explicitly chooses an equivalent of Approve and continue now; in that case set next_action: continue_now before continuing
• mode mapping for next_skill_recommended:
  • recommended_mode=swarm → pulse:swarming
  • recommended_mode=single-worker → pulse:executing

On rejection:

• capture concern category and route back precisely (shape/current-work/approach/beads)

Use final approval template/options in references/runtime-appendix.md.

Execution Mode Compatibility

Validating must support both downstream modes:

• Swarm mode: all validations complete, recommend pulse:swarming as next skill and default to next_action: manual_invoke
• Single-worker mode: all validations complete, recommend pulse:executing as next skill and default to next_action: manual_invoke

Only switch to next_action: continue_now when the user explicitly asks to keep going in the same context. The validation standard is identical in both modes.

Lightweight Mode

Allowed only for confirmed LOW-risk direct/small single-slice work:

• abbreviated reality + feasibility check
• skip spikes
• run bv --robot-suggest when beads exist
• still require Gate 3 explicit approval

If uncertain, run full mode.

Context Budget

If context exceeds 65%, write a validating-owned handoff using the shared envelope and register it in .pulse/handoffs/manifest.json.

Red Flags

• any execution before Gate 3 approval
• skipping orientation on resume
• validating while shape approval is not APPROVED
• validating when shape artifact and .pulse/STATE.md disagree
• running structural checker before reality/feasibility/schema pass
• attempting a 4th structural iteration
• missing or non-decisive HIGH-risk spike outcomes
• continuing after a NO spike
• unresolved CRITICAL bead-review flags at approval time