Skill orientation

Use this panel to select and sequence the skill. The canonical source follows below.

Purpose: Run the selected work-type skill against the immutable route.
Category and sequence: Lifecycle; sequence 5
Primary use: Execute The Routed Engineering Workflow, Run The Durable Checkpoint Loop, Coordinate The Selected Work Type Skill
Required gates: Goal
Conditional gates: code-intelligence for code, architecture, debugging, refactor, or cross-file claims; trajectory for multi-checkpoint or agentic work; delivery-primary validation selected by the immutable route; Red Zone approval before consequential actions
Red Zone triggers: No skill-specific triggers are declared
Next route: Prove and Govern
Source identifier: skills/valdris-execute-workflow/SKILL.md

Canonical pathskills/valdris-execute-workflow/SKILL.mdRevision69bab1cInspect source

Valdris Execute Workflow

This skill owns Workflow Execution. It coordinates the selected work-type skill; it does not reclassify the request.

Deterministic flow

Run the intake, classification, route, goal, foundation, and applicable assurance preflight gates. Stop before mutation if a required pre-implementation artifact fails.
Read run/route.json. Use its delivery primary as the engineering workflow and only its declared supporting skills. Do not select a more convenient primary or exceed four supporting skills.
Run code intelligence before codebase, architecture, debugging, refactor, or cross-file claims. Cite design anchors and refresh the graph whenever a bound source input changes.
Inspect the current goal revision, latest checkpoint, remaining risk, budgets, forbidden actions, and stopping conditions. Choose the smallest bounded slice that can produce new evidence.
Execute the selected workflow in its declared order. Use failing tests at pre-agreed seams, minimal implementation, task-specific validation, and refactoring under passing checks.
Record code, tests, evals, trajectory, migrations, smoke preparation, and affected assurance artifacts. Emit truthful events in Live Run mode.
Before any Red Zone action, request scoped human approval and stop. Never treat a user desire for completion as approval for credentials, deployment, billing, customer data, destructive migration, traffic, or provider changes.
Verify the slice, then advance goal/goal.json only with goal-transition.mjs --expected-revision <n>. On stale revision, reload instead of overwriting. Continue only while the next slice is within the immutable budget and adds evidence.
Stop as passed, blocked, failed, or escalated. Repeated failure without new evidence triggers RCA or self-heal, not an infinite loop.

Completion criterion

Execution is complete only when the requested behavior is implemented; acceptance and regression evidence exists; affected controls are updated; every recorded checkpoint transition is revision-safe; no budget or authority boundary was crossed; and the goal is ready for finish-line proof.

Then hand off to $valdris-prove-govern.