spec-kitty.plan
Create an implementation plan
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 2
- Forks
- 0
01 前言 Front Matter
- name
- spec-kitty.plan
- description
- Create an implementation plan
- user-invocable
- true
03 SKILL.md 正文
Startup Upgrade Check
Run this at most once per active agent session before the first Spec Kitty command workflow.
If you already ran spec-kitty upgrade --agent-check --json in this session, reuse that result and skip this block.
Do not run or announce an upgrade check again for later Spec Kitty commands in the same session.
Otherwise, before continuing, run:
spec-kitty upgrade --agent-check --json
If JSON action is none, continue.
If action is auto_upgrade, run upgrade_command before continuing. If it fails, tell the user and continue with the current Spec Kitty version.
If action is guidance, show upgrade_note briefly, then continue.
If action is prompt, ask the user with the host-native question UI when available:
Spec Kitty {latest_version} is available. You are on {installed_version}. Upgrade now?
Use these choices:
- Upgrade now (recommended) - record
upgrade_now, runupgrade_command, then continue. - Always keep me up to date - record
always, runupgrade_command, then continue. - Not now - record
not_now, then continue. - Never ask again - record
never_ask, then continue.
Record the selected choice before continuing:
spec-kitty upgrade --agent-choice <upgrade_now|always|not_now|never_ask> --agent-latest <latest_version> --json
If no host-native question UI is available, present the same four choices in plain text and wait for the user.
In non-interactive hosts, choose not_now and continue.
/spec-kitty.plan - Create Implementation Plan
Version: 0.11.0+
REASONS Guidance — Plan
While composing the plan, fill or update:
- Approach — chosen strategy and rejected alternatives with rationale.
- Structure — code surfaces, components, dependencies, ownership boundaries.
Link to source artifacts (spec, contracts) instead of duplicating them.
📍 WORKING DIRECTORY: Stay in the repository root checkout
IMPORTANT: Plan works in the repository root checkout. NO worktrees created.
# Run from project root (same directory as /spec-kitty.specify):
# You should already be here if you just ran /spec-kitty.specify
# Creates:
# - kitty-specs/<mission_slug>/plan.md → In repository root checkout
# (the NNN- prefix in the directory listing is display-only metadata)
# - Commits to target branch
# - NO worktrees created
Do NOT cd anywhere. Stay in the repository root checkout.
Mission Handle Rule
/spec-kitty.plan operates on an existing mission, so use --mission <handle>
when the CLI needs a mission selector.
<handle>can be the mission'smission_id(ULID),mid8(first 8 chars of the ULID), ormission_slug.- Prefer
mission_idormid8when the repo has multiple similarly named missions. - The resolver disambiguates by
mission_idand returns a structuredMISSION_AMBIGUOUS_SELECTORerror on ambiguity — there is no silent fallback.
User Input
The content of the user's message that invoked this skill (everything after the skill invocation token, e.g. after /spec-kitty.<command> or $spec-kitty.<command>) is the User Input referenced elsewhere in these instructions.
You MUST consider this user input before proceeding (if not empty).
Commit Boundary (issue #846)
/spec-kitty.plan will refuse to advance the plan phase unless two
gates pass:
-
Entry gate.
spec.mdmust already be both committed (tracked + present at HEAD) and substantive (at least one populatedFR-###row — not just template placeholders). If either check fails, the CLI returnsphase_complete=falsewith ablocked_reasonnaming "committed AND substantive" and does not create or commitplan.md. -
Exit gate.
plan.mdis only auto-committed when its Technical Context section contains a realLanguage/Versionvalue (and at least one peer field) — not the[e.g., …]/[NEEDS CLARIFICATION …]placeholders.- The FIRST
setup-plancall after the entry gate passes scaffoldsplan.mdfrom the template. This is a non-error state: the CLI returnsresult: "success"withscaffold_only: trueandphase_complete: false— it means "plan.md is ready for you to populate", not a failure. Do not treat this call as blocked; proceed to fill in the Technical Context. - Once
plan.mdhas been edited but its Technical Context is still not substantive, the CLI returnsresult: "blocked"withphase_complete: falseand a populated-but-insufficientblocked_reasonnaming the missing field(s) — populate the section and re-run.
- The FIRST
Section presence is the only signal — adding arbitrary prose without the required structural rows does not count as substantive (no byte-length escape hatch).
To advance: populate the Technical Context with real values, then re-run
spec-kitty agent mission setup-plan --mission <mission-slug> --json. The substantive plan will be
auto-committed and phase_complete will report true.
Issue-Matrix Approval Heads-Up (non-gating, #3469)
If plan.md/research.md cite a GitHub issue number (#NNNN), a bare/unmarked
reference will later require an issue-matrix row before its owning work package can be
approved. A context-only citation (e.g. Follow-up:, see #, parent, epic) or a
PR/commit reference (PR #NNNN, a /pull/NNNN URL) is non-gating and needs no row; if
an issue genuinely owes the mission no work, it can later be recorded with the
not-applicable verdict. This is informational only — it does not gate
/spec-kitty.plan.
Branch Strategy Confirmation (MANDATORY)
Before asking planning questions or generating artifacts, you must make the branch contract explicit.
- Never describe the landing branch vaguely. Always name the actual branch value.
- If the user says the mission should land somewhere else, stop and resolve that before writing
plan.md. - You must repeat the branch contract twice during this command:
- immediately after parsing
setup-plan --json - again in the final report before suggesting
/spec-kitty.tasks
- immediately after parsing
Charter Context Bootstrap (required)
Before planning interrogation, load charter context for this action:
spec-kitty charter context --action plan --json
- If JSON
modeisbootstrap, apply JSONtextas first-run governance context and follow referenced docs as needed. - If JSON
modeiscompact, continue with condensed governance context.
Visual Communication (recommended)
Apply the visual doctrine for non-trivial architecture, data/control flow,
boundaries, migrations, or risky interactions. Load spk-practice-show-me and
use the smallest diagram
that materially reduces prose. Prefer Mermaid in Markdown; use PlantUML when
richer layout, mature C4 support, or its DSL earns the added rendering cost.
Apply C4 progressive zoom for architecture and stop at the first level that
answers the planning question. Text contracts and plan decisions remain
authoritative; do not generate speculative diagrams.
Location Check (0.11.0+)
This command runs in the repository root checkout, not in a worktree.
- Resolve branch context from deterministic JSON output, not from
meta.jsoninspection:- Run
spec-kitty agent mission setup-plan --mission <mission-slug> --json - Use
current_branch,target_branch/base_branch, andplanning_base_branch/merge_target_branch(plus uppercase aliases) from that payload - Use
branch_matches_targetfrom that payload to detect branch mismatch; do not probe branch state manually inside the prompt
- Run
- Planning artifacts live in
kitty-specs/<mission_slug>/(theNNN-prefix is display-only metadata) - The plan template is committed to the target branch after generation
Path reference rule: When you mention directories or files, provide either the absolute path or a path relative to the project root (for example, kitty-specs/<mission>/tasks/). Never refer to a folder by name alone.
Agent Context Files (do not mutate)
This command does not update agent-specific context files.
- Do not search for or mutate
CLAUDE.md,AGENTS.md, or similar agent-specific files as part of/spec-kitty.plan. - Do not hunt for updater scripts or imaginary
spec-kitty agent context updatecommands. No supported context-update command exists in this release. - Planning outputs are the mission planning artifacts only:
plan.mdresearch.mddata-model.mdcontracts/quickstart.mdoccurrence_map.yamlwhen bulk-edit planning applies
Decision Moment Protocol
Before asking any clarifying question during plan elaboration, you MUST:
-
Run
spec-kitty agent decision opento mint a decision_id:spec-kitty agent decision open \ --mission <mission-slug> \ --flow plan \ --slot-key plan.<section>.<question-slug> \ --input-key <snake_case_key> \ --question "<question text>" \ [--options '["option1","option2","Other"]']Capture the returned
decision_idfrom the JSON output. -
Ask the question to the user in chat.
-
After the user answers, run exactly one of:
- Resolved:
spec-kitty agent decision resolve <decision_id> --mission <slug> --final-answer "<answer>" - Deferred:
spec-kitty agent decision defer <decision_id> --mission <slug> --rationale "<reason>" - Canceled:
spec-kitty agent decision cancel <decision_id> --mission <slug> --rationale "<reason>"
- Resolved:
-
When deferring, write the inline marker into
plan.md:[NEEDS CLARIFICATION: <brief description>] <!-- decision_id: <decision_id> --> -
Before finishing this command, run:
spec-kitty agent decision verify --mission <slug>Resolve all findings (DEFERRED_WITHOUT_MARKER,MARKER_WITHOUT_DECISION,STALE_MARKER) before proceeding.
Important constraints:
--slot-keyformat:plan.<section>.<question-slug>(e.g.,plan.architecture.db-choice).--input-keyis the snake_case programmatic key (e.g.,db_choice).- The
decision_idon the wire is a plain ULID (26 chars). TheDM-prefix appears only in artifact filenames, not in CLI arguments. - The verifier cross-checks
[NEEDS CLARIFICATION: …] <!-- decision_id: <id> -->sentinels inplan.mdagainst the decisions index and exits non-zero on drift. - Widening is represented by the CLI/SaaS widen flow; if that flow returns
canonical thread metadata, it must be recorded as
DecisionPointWidened. - Local-only; no SaaS calls needed.
Planning Interrogation (mandatory)
Before executing any scripts or generating artifacts you must interrogate the specification and stakeholders.
-
Scope proportionality (CRITICAL): FIRST, assess the request's complexity from the spec:
- Trivial/Test Features (hello world, simple static pages, basic demos): Ask 1-2 questions maximum about tech stack preference, then proceed with sensible defaults
- Simple Features (small components, minor API additions): Ask 2-3 questions about tech choices and constraints
- Complex Features (new subsystems, multi-component features): Ask 3-5 questions covering architecture, NFRs, integrations
- Platform/Critical Features (core infrastructure, security, payments): Full interrogation with 5+ questions
-
Scenario-to-design handoff: For non-trivial features, anchor planning questions in the concrete user flows from the spec rather than generic architecture preferences.
-
Domain rule follow-through: When the spec implies approvals, lifecycle states, irreversible operations, or business-critical validations, ask 1-2 targeted questions about invariants, transitions, atomicity, and externally visible events or integrations. Skip this for trivial features.
-
User signals to reduce questioning: If the user says "use defaults", "just make it simple", "skip to implementation", "vanilla HTML/CSS/JS" - recognize these as signals to minimize planning questions and use standard approaches.
-
First response rule:
- For TRIVIAL features: Ask ONE tech stack question, then if answer is simple (e.g., "vanilla HTML"), proceed directly to plan generation
- For other features: Ask a single architecture question tied to the riskiest scenario, rule, or lifecycle constraint and end with
WAITING_FOR_PLANNING_INPUT
-
If the user has not provided plan context, keep interrogating with one question at a time.
-
Conversational cadence: After each reply, assess if you have SUFFICIENT context for this mission's scope. For trivial features, knowing the basic stack is enough. Only continue if critical unknowns remain.
Planning requirements (scale to complexity):
- Maintain a Planning Questions table internally covering questions appropriate to the mission's complexity (1-2 for trivial, up to 5+ for platform-level). Track columns
#,Question,Why it matters, andCurrent insight. Do not render this table to the user. - For trivial features, standard practices are acceptable (vanilla HTML, simple file structure, no build tools). Only probe if the user's request suggests otherwise.
- When you have sufficient context for the scope, summarize into an Engineering Alignment note and confirm. Include invariant, state-transition, or event assumptions when they materially affect the design.
- If user explicitly asks to skip questions or use defaults, acknowledge and proceed with best practices for that kind of work.
Bulk-Edit Check (if applicable)
If this mission is marked change_mode: bulk_edit in meta.json — or if the
spec describes renaming the same string (identifier, path, key, label, term)
across many files — load the spk-practice-bulk-edit skill and
follow it. You will produce kitty-specs/<mission>/occurrence_map.yaml
alongside the other planning artifacts. Every one of the 8 standard categories
(code_symbols, import_paths, filesystem_paths, serialized_keys, cli_commands,
user_facing_strings, tests_fixtures, logs_telemetry) must have an explicit
action. Without that artifact, the implement command will refuse to start
the first WP.
If the mission is not a bulk edit, skip this step.
Supply-Chain Security & Adversarial Evidence (Planning)
Apply this section whenever the plan adds, upgrades, or removes a dependency, in any ecosystem (npm/yarn/pnpm, pip/uv, Maven/Gradle, etc.).
- Security checks: Reference
DIRECTIVE_051and thesupply-chain-install-safetytactic. Planning output (Technical Context and/orresearch.md) must surface a result for every control the directive names for any dependency decision — this mirrors thesupply_chain_security_checkstep already present in theplanstep contract. Use the matching ecosystem toolguide for the exact commands. - Advisory posture: This is advisory in v1 — it does not add a new blocking gate to the Commit Boundary gates above — but an unexamined default is a gap in the plan, not a pass. Silence is not compliance.
- Adversarial evidence (mandatory for plan/research): When a security-impacting dependency decision is made, run (or explicitly document deferral of) an adversarial-squad challenge pass before claiming plan readiness, per the
adversarial-squad-deploymentprocedure. Record each contested finding's disposition inresearch.mdper that procedure's findings-disposition contract (load it withspec-kitty charter context --include procedure:adversarial-squad-deployment). No contested finding may be silently dropped.
Outline
-
Check planning discovery status:
- If any planning questions remain unanswered or the user has not confirmed the Engineering Alignment summary, stay in the one-question cadence, capture the user's response, update your internal table, and end with
WAITING_FOR_PLANNING_INPUT. Do not surface the table. Do not run the setup command yet. - Once every planning question has a concrete answer and the alignment summary is confirmed by the user, continue.
- If any planning questions remain unanswered or the user has not confirmed the Engineering Alignment summary, stay in the one-question cadence, capture the user's response, update your internal table, and end with
-
Resolve mission context deterministically (CRITICAL - prevents wrong mission selection):
- Prefer an explicit mission slug from user direction or from the current directory path (
kitty-specs/<mission-slug>/...) - The resolver requires
--mission— always pass an explicit handle. Runningsetup-planwithout--missionreturnsPLAN_CONTEXT_UNRESOLVEDeven when exactly one mission is present. - Resolve the handle first:
spec-kitty agent context resolve --action plan --mission <handle> --json, then pass the resolved slug tosetup-plan - If the context resolve call returns an ambiguity error with
available_missions, stop and pick one explicit mission slug before continuing
- Prefer an explicit mission slug from user direction or from the current directory path (
-
Setup: Run
spec-kitty agent mission setup-plan --mission <mission-slug> --jsonfrom the repository root and parse JSON for:result: "success" or error messagescaffold_only:trueonly on the first happy-path scaffold write (plan.md freshly copied from the template, untouched). This isresult: "success"and NOT an error — populate the Technical Context and re-runsetup-planto commit.phase_completestaysfalseuntil then.mission_slug: Resolved mission slugspec_file: Absolute path to resolved spec.mdplan_file: Absolute path to the created plan.mdmission_dir: Absolute path to the mission directory (feature_diris a legacy alias key)current_branch: branch checked out when planning startedtarget_branch/base_branch(deterministic branch contract for downstream commands)planning_base_branch/merge_target_branch: explicit aliases for planning and merge intentbranch_strategy_summary: canonical sentence describing the branch strategy
Before proceeding, explicitly state to the user:
- Current branch at plan start
- Intended planning/base branch
- Final merge target for completed changes
- Whether
branch_matches_targetsays the current branch matches that intended target
Example:
# Resolve the active mission handle, then pass it to setup-plan. # The --mission flag accepts mission_id (ULID), mid8 (first 8 chars), or mission_slug. # The resolver disambiguates by mission_id; ambiguous handles become structured errors. # --action is required for context resolve; use the action matching this step. spec-kitty agent context resolve --action plan --mission <handle> --json spec-kitty agent mission setup-plan --mission <handle> --jsonError handling: If the command fails with "Cannot detect mission", "Multiple missions found", or
MISSION_AMBIGUOUS_SELECTOR, pass an unambiguous handle — themission_idor its 8-char prefixmid8always disambiguates. -
Load context: Read
spec_filefrom setup-plan JSON output and.kittify/charter/charter.mdif it exists. If the charter file is missing, skip Charter Check and note that it is absent. Load IMPL_PLAN template (already copied). -
Execute plan workflow: Follow the structure in IMPL_PLAN template, using the validated planning answers as ground truth:
- Update Technical Context with explicit statements from the user or discovery research; mark
[NEEDS CLARIFICATION: …] <!-- decision_id: <id> -->only when the user deliberately postpones a decision (calldecision deferbefore writing each such marker) - If a charter exists, fill Charter Check section from it and challenge any conflicts directly with the user. If no charter exists, mark the section as skipped.
- Evaluate gates (ERROR if violations unjustified or questions remain unanswered)
- Phase 0: Generate research.md — extend it if the discovery step already created one; do not replace existing content (commission research to resolve every outstanding clarification, prioritizing unresolved domain rules, lifecycle questions, and event/integration behavior before generic tech comparisons)
- Phase 1: Generate data-model.md, contracts/, quickstart.md based on confirmed intent; when applicable, capture entities/value objects, invariants, state transitions, and externally visible events in the design artifacts
- Re-evaluate Charter Check post-design, asking the user to resolve new gaps before proceeding
- Update Technical Context with explicit statements from the user or discovery research; mark
-
STOP and report: This command ends after Phase 1 planning. Report branch, IMPL_PLAN path, and generated artifacts (including the Implementation Concern Map if present).
⚠️ CRITICAL: DO NOT proceed to task generation! The user must explicitly run
/spec-kitty.tasksto translate implementation concerns fromplan.mdinto executable work packages. Your job is COMPLETE after reporting the planning artifacts.
Phases
Phase 0: Outline & Research
-
Extract unknowns from Technical Context above:
- For each NEEDS CLARIFICATION → research task
- For each unresolved rule, invariant, lifecycle edge, or event/integration ambiguity → domain research task
- For each dependency → best practices task
- For each integration → patterns task
-
Generate and dispatch research agents:
For each unknown in Technical Context: Task: "Research {unknown} for {mission context}" For each unresolved domain rule: Task: "Clarify invariant, state transition, or event behavior for {mission context}" For each technology choice: Task: "Find best practices for {tech} in {domain}" -
Consolidate findings in
research.md: extend the file if discovery already created it; do not replace existing content. Use format:- Decision: [what was chosen]
- Rationale: [why chosen]
- Alternatives considered: [what else evaluated]
Output: research.md with all NEEDS CLARIFICATION resolved
Phase 1: Design & Contracts
Prerequisites: research.md complete
-
Extract entities from mission spec →
data-model.md:- Entity name, fields, relationships
- Validation rules from requirements
- Invariants or atomicity boundaries if applicable
- State transitions if applicable
-
Generate API contracts from functional requirements:
- For each user action → endpoint
- For each externally visible event, webhook, or integration callback → contract or payload shape when applicable
- Use standard REST/GraphQL patterns
- Output OpenAPI/GraphQL schema to
/contracts/ - If the Mission defines no interfaces (test remediation, refactor), do not
create an empty
contracts/. Instead add"contracts": "none"and a non-empty"contracts_rationale"to the Mission'smeta.json, so strictspec-kitty acceptwaives thecontracts/requirement.
Output: data-model.md, /contracts/*, quickstart.md
Key rules
- Use absolute paths
- ERROR on gate failures or unresolved clarifications
⛔ MANDATORY STOP POINT
This command is COMPLETE after generating planning artifacts.
After reporting:
plan.mdpathresearch.mdpath (if generated)data-model.mdpath (if generated)contracts/contents (if generated)
YOU MUST STOP HERE.
Do NOT:
- ❌ Generate
tasks.md - ❌ Create work package (WP) files
- ❌ Create
tasks/subdirectories - ❌ Proceed to implementation
/spec-kitty.tasks translates implementation concerns from plan.md into executable work packages. The user will run it when they are ready.
Next suggested command: /spec-kitty.tasks (user must invoke this explicitly)
04 同仓库更多 Skills
kitty-semi-auto
Use to take a tracker issue or a coherent issue-set through a governed delivery run that works autonomously but pauses to escalate the operator's decisions. Runs BOTH ways — in this session (local) or handed to a fresh cloud runner (cloud) — from one set of standing orders covering orientation, code grounding first, Mission-vs-Ops, the pause-and-escalate contract, brownfield discipline, and the canonical closeout to a draft PR the operator merges. Triggers: semi-auto, semi-auto dispatch, run this semi-automatically, deliver these issues semi-auto, hand this to a cloud runner, dispatch semi-auto.
kitty-report-debrief
Use when a maintainer asks what happened or what is open ("what landed since Friday", "WTF happened with X", a milestone overview). Collects the facts from GitHub, writes a short branded debrief for named readers, and hands the PDF back without committing it. Triggers: what landed since, WTF happened, debrief the weekend, milestone overview, exec overview of open issues, write up what shipped.
kitty-mission-from-issue
Use to take one or more GitHub issues through a full governed Mission (spec, plan, tasks, implementation, review) to a green pull request flipped to ready for review, which a maintainer then merges. Triggers: mission from issue, run a mission for this issue, drive issue to PR.
kitty-land-pr
Use when a ready-for-review pull request needs a maintainer to get it ready to merge. Rebases it, fixes and reviews it, updates its description, changelog and docs, and hands it back without merging by hand. Triggers: land this PR, landing pass, land PR.
kitty-issue-triage
Use to triage a batch of open GitHub issues on spec-kitty/spec-kitty (default, those opened in the last 7 days). Sets type, priority, domain, parent and milestone, and reports stale, duplicate or clustered issues for an operator decision. Triggers: triage issues, triage the tracker, triage new issues.
spk-team-tracker
暂无简介