| name | planning |
| description | Produce deterministic execution plans with story artifacts ready for workflow-driven delivery. Use when tasks are multi-step, high-risk, ambiguous, or span multiple files/services. |
Planning
Use this skill to produce execution-ready plans before implementation.
Planning Loop
- Explore relevant codepaths, docs, and constraints first.
- Enumerate ambiguous branches and unknowns explicitly.
- Interview the plan across the full design tree:
- Challenge assumptions directly, surface blind spots/edge cases/sequencing risks, and avoid passive transcription.
- Ask one question at a time and wait for feedback before continuing; only batch questions when they are fully independent, and limit each batch to at most three questions.
- Resolve decision dependencies one-by-one; do not advance while upstream decisions remain open.
- If a question can be answered by exploring the codebase, investigate first and only ask when uncertainty remains.
- For each question you ask me, provide a recommended answer first.
- Base recommendations on concrete codebase/docs evidence when available; when evidence is unavailable, state the assumption explicitly before requesting confirmation or correction.
- Continue until all implementation-critical decisions are resolved (decisions that materially change execution steps, ordering, or scope), no unresolved upstream dependencies remain, and no open questions remain for final plan output.
- Once all open questions are resolved and before presenting final plan output, apply the canonical critique gate defined in
AGENTS.md.
Execution Readiness and Guardrails
- Do not start execution with unresolved decision branches.
- Do not leave implementation-critical steps as
TBD or "decide later."
- Ensure execution steps are ordered, concrete, and testable.
- Ensure rollback/mitigation approach is clear before execution.
Plan Output Requirements
When producing or validating a plan, include only actionable execution instructions:
- Objective
- Deterministic story plan artifact:
- Format implementation work as story blocks that execution agents can follow without reinterpretation.
- Each story block must include: stable story key, dependencies, numbered implementation steps, acceptance checks, and rollback/mitigation notes.
- Keep story keys machine-parseable and execution order topologically valid.
- Execution handoff constraints:
- Declare sequencing constraints and concurrency boundaries required by execution workflow.
- Identify vertical integration checkpoints that execution must preserve as feedback anchors.
- Include the exact persisted plan artifact path in the handoff and require execution to consume that exact path (no default-path inference).
- Verification/acceptance checks
All open questions must be resolved during the planning conversation before presenting final plan output.
Do not include meta-commentary, decision-making history, tradeoff history, or rationale narrative in final plan output.
Execution Handoff
Once the final plan artifact is accepted, load the execution-workflow skill to drive story delivery.
Persist the accepted plan artifact using the canonical location and naming conventions defined by the agentic-projects skill before handoff, and include that exact plan path as an explicit handoff field for execution.