| name | plan-task-standards |
| description | Self-contained CatHerder method and standards for plans and tasks: planning vs execution, execution loop, prompt precedence, persistence; plus prompt/spec/implementation/review/follow-up files, frontmatter, statuses, timestamps, the open-questions gate, and validation. Use when planning, creating, reviewing, executing, or validating planNNN folders, or fixing plan/task format drift. |
| metadata | {"version":"1.6"} |
Plan & Task Standards
CatHerder standards for plan folders and task files.
This file is a router. Read only the section for your job, then open the one
reference it points to. You do not need to read everything.
This skill targets worker agents. An orchestrator (human or deterministic
driver) directs sequencing and dispatches work; it is not bound by this skill
(see references/method.md).
Posture
CatHerder is deliberately strict on structure, loose on content and process.
- Strict (validator-enforced): frontmatter keys, status vocabularies,
required sections, the implementation task-table shape, the open-questions
gate, and timestamps. These are machine-checked so a human+agent team can
trust a plan's state without re-reading it.
- Loose (guidance only): what you write in each section, how you split work
into tasks, how granular tasks are, and how you verify. These are judgement
calls, not rules.
The structure exists to make the process lightweight and reliable, not to add
ceremony. It is a standard for the artifacts, not a rigid procedure for how you
must work.
Canonical Defaults
| Directory | Purpose |
|---|
.agents/ | Agent-facing assets: agent definitions, instruction files, skills, and MCP configuration |
.catherder/ | CatHerder process artifacts: plans and decisions |
Semantic names used after defaults are established:
| Semantic name | Default location |
|---|
| agent instructions | .agents/ |
| plans folder | .catherder/plans/ |
| decisions folder | .catherder/decisions/ |
| project CatHerder guidance | .agents/catherder.project.instructions.md |
These are canonical defaults. Projects may vary when they have a reason.
Find Your Job
Each row below is a separate dispatch. Do the one you were dispatched for,
then stop — do not start the next stage unless your dispatch includes it.
| You are about to… | Role | Read |
|---|
| Understand the CatHerder method (always) | any | references/method.md |
| Write the prompt and spec (intent) | planner | references/role-planner.md |
| Turn a spec into an implementation with tasks (a separate dispatch) | planner | references/role-planner.md |
| Execute a task | executor | references/role-executor.md |
| Review a spec/implementation, or verify a task | reviewer | references/role-reviewer.md |
| Validate or fix file format/drift | any | run the validator (below) |
The full rule set is references/plan-task-spec-2026-05.md; open it only when a
role file points you there.
This skill is self-contained. If the project has project CatHerder guidance,
read it and follow its guidance (identity, phase, git workflow, build/test).
Plan Folder Files
planNNN-short-description/
planNNN-prompt.md # original request
planNNN-spec.md # intent: what & why
planNNN-implementation.md # executable plan + tasks
planNNN-post-followup.md # post-completion tracking (optional)
data/ tasks/ review/
Lifecycle: prompt → spec → implementation → (review) → completed → (follow-up).
Plans live in the plans folder. Decisions (ADRs) live in the decisions folder.
Quick Rules (all roles)
- Status, timestamps, and
type live in frontmatter (name is not used).
- Open-questions gate: a spec is
ready only when every open question is
resolved. Resolving questions is required for ready, not a green light to
build the implementation — that still needs a dispatch.
- A task is
completed only after verification passed and review accepted.
Status values, timestamp format, required sections, and the task-table shape
are in references/plan-task-spec-2026-05.md.
Nested Projects
A nested project (e.g. a Git submodule) is its own CatHerder project with its
own skills/plans — see references/method.md.
Scripts
| Script | Purpose | Usage | Exit Codes |
|---|
scripts/validate.sh | Run the validator via the workspace venv. Report-only. | bash scripts/validate.sh path/to/planNNN-… | 0=clean 1=issues 2=usage |
scripts/validate.py | Validate structure, frontmatter, status, timestamps, task table, open-questions gate. Emits JSON. | python scripts/validate.py path/to/planNNN-… | 0=clean 1=issues 2=usage/config |
validate.py needs python-frontmatter>=1.1.0.
References Map
- Method (behavioural rules):
references/method.md
- Roles (start here):
references/role-planner.md,
references/role-executor.md, references/role-reviewer.md
- Canonical rules + status lifecycle:
references/plan-task-spec-2026-05.md
- Templates:
PROMPT_TEMPLATE.md, SPEC_TEMPLATE.md,
IMPLEMENTATION_TEMPLATE.md, TASK_TEMPLATE.md, REVIEW_TEMPLATE.md,
FOLLOWUP_TEMPLATE.md (all under references/)
- Validator rulebook:
references/validation-rules-2026-05.md
- Trigger/non-trigger prompts:
references/test-prompts-2026-05.md
- Project adoption:
references/AGENTS_EXAMPLE.md
Project Process Rules
This skill is self-contained for the CatHerder method and plan/task format.
Project-specific rules (identity, phase, git workflow, build/test) live in
project CatHerder guidance. If that file is present, read and follow it; it may
add constraints but must not override this skill's method. New projects adopt
the skill via references/AGENTS_EXAMPLE.md.
Scope
- In scope: plan folder layout, prompt/spec/implementation/review/follow-up
format, lifecycle, frontmatter, status values, timestamps, the
open-questions gate, structural validation.
- Out of scope: skill file format (see
skill-file-standards), AGENTS.md
standards, and project source-code conventions.
