| name | star-plan-decomposer |
| disable-model-invocation | true |
| description | Decompose an existing research plan (written by star-plan-coach and living under metds/plans/) into concrete, executable sub-plans. Reads the parent plan, picks a decomposition axis (milestone / component / claim→experiment), then auto-drafts one execution sub-plan per unit — objective, dependencies, task breakdown, deliverables, and done-criteria — writing each to metds/plans/ under a hierarchical numeric prefix and linking it back to the parent. Supports arbitrary decomposition depth. Use when the user runs /star-plan-decomposer, or wants to break down / flesh out the concrete execution details of a plan, turn a plan's method or milestones into actionable tasks, or split a plan into sub-plans. Bilingual (en/zh). |
Research Plan Analyse — plan decomposer
Match the user's language. For Chinese dialogue, read SKILL_zh.md in full before acting and follow it as the localized instructions; load other *_zh.md resources when referenced. Otherwise, follow this file and load unsuffixed resources. If SKILL_zh.md conflicts with this file, this SKILL.md is authoritative.
Invocation: /star-plan-decomposer PLAN_NAME, where PLAN_NAME is a slug (open-vocab-det-seg), a numeric prefix (0), or a filename (0_open-vocab-det-seg_plan.md).
Shared conventions. Read docs/mds/star-workflow/research-workflow-conventions.md (Chinese: research-workflow-conventions.zh-CN.md) before acting: §1 git, §2 the STOP line, §3 .env runtime, §4 real dates, §5 plan-name resolution, §6 delegation, §7 dialogue, §8 the artifact registry, §9 project layout. It is the baseline every STAR skill shares; this file states what is specific to this one, and wins wherever it is stricter.
Role
You take a strategic research plan and turn it into executable sub-plans. The sibling skill star-plan-coach produces the strategy (one root plan: problem → related work → method → experiments → risks → milestones). This skill produces the execution: it splits the plan's concrete implementation into smaller sub-plans, each with steps a researcher can actually run and verify.
You decompose, you do not re-strategize. The parent plan already holds the thinking — pull execution detail out of it; do not re-derive the research question, novelty, or method from scratch.
Core Principles
- Decompose, don't re-strategize. The parent is the source of truth for why and what. Your job is how: sub-goals, ordered steps, dependencies, deliverables, and a check that proves each is done. If you find yourself questioning the research question or method, stop — that belongs in
star-plan-coach, not here.
- Confirm the shape, then auto-draft the content. Two decisions go through the user (one question at a time, each with a recommendation): the decomposition axis, then the sub-plan list. After those are confirmed, draft each sub-plan's content autonomously by pulling from the parent. Mark genuine gaps
[TBD]; ask a targeted follow-up only when a step is undecidable without the user. Do not re-elicit detail the parent already records.
- Incremental writes. Write each sub-plan file the moment it is drafted. Prefer more file writes over leaving results in chat — chats end, files do not.
- Every sub-plan is verifiable. A sub-plan is not done until it has concrete, verb-specific steps, a done-criterion (a test / metric / output that proves completion), and deliverables placed per the project layout (
datas/, inits/, code/, wkdrs/<run>, and tasks/<plan-name>/ for the plan's own tool scripts). This mirrors the project's Goal-Driven Execution and Verification rules.
- Traceability both ways. Every sub-plan names the root section or claim it traces to (
traces_to). The parent gets a ## Sub-plans index and a children: frontmatter list. The numeric prefix orders the tree for humans; the frontmatter parent: field is the authoritative link.
- Dependencies are first-class, not prose-only. Each sub-plan carries a
depends_on: frontmatter list — the sibling prefixes that must finish before it can start. This is the machine-readable order the executor and star-flow-status consume to answer "what's runnable next". Keep it a DAG (no cycles) and consistent with the ## Sub-plans index order.
Naming Convention (summary)
Filenames are <prefix>_<slug>_plan.md. The prefix is a string of decimal digits; its length equals the plan's depth in the tree.
- To decompose a plan with prefix
P, its sub-plans get prefix P with one more digit appended = the child's 0-based index: 0_ → 00_ 01_ 02_ …; 00_ → 000_ 001_ …; 3_ → 30_ 31_ ….
- Parent = drop the last digit. Level = prefix length. Max 10 siblings per node (indices 0–9).
Full rule, worked tree, and edge cases: references/naming_convention.md.
Workflow
Step 0: Resolve the target plan
- Interpret
PLAN_NAME: match it against metds/plans/*_plan.md by slug, by numeric prefix, or by full filename.
- If no argument was given, or the match is ambiguous, list the available plans (prefix + slug + one-line title) and ask which one (one question).
- Read the resolved plan in full.
Step 1: Assess readiness
First, check whether this plan has already been decomposed. Scan metds/plans/ for files whose parent: is the target — equivalently, whose prefix is the target's prefix plus one digit. If any exist, decomposition is already partial or complete, and Steps 2–4 would overwrite files that may carry hand edits, a ## Revision History, or execution state. Report what was found (prefix, slug, exec_status, and whether the parent's ## Sub-plans / children: already list them) and offer:
- Repair the parent index only (recommended when the existing children look complete) — skip Steps 2–4 and go straight to Step 5, deriving the index from the child files themselves. Nothing is written to the children.
- Add new units alongside them — leave the existing files untouched, number new units from the next free index, and run Steps 2–4 for those only; Step 5 then merges old and new.
- Re-decompose from scratch — Steps 2–4 as normal, but confirm each overwrite file-by-file, and never overwrite a child carrying
## Revision History or a non-empty exec_runs without naming exactly what would be lost.
Check the root's finalized: — the one signal that a strategy plan is ready to consume (star-plan-coach sets it only when all six sections are done/skipped and the rubric passed, and clears it whenever a section reopens). Not finalized → read its status map and body, name which sections are pending/in_progress or [TBD]-ridden (especially method and milestones), and tell the user that decomposition will be shallow, and offer: decompose anyway (gaps become [TBD] in sub-plans) / go back to star-plan-coach to finish the parent first. Respect the choice.
If the target itself carries execution evidence (exec_runs non-empty, or exec_status beyond pending), pause before splitting: decomposition turns an executed leaf into an internal node — its exec_status / exec_runs freeze as history, star-flow-status stops counting it as an executable leaf, and its wkdrs/ runs stay attached to a node no executor revisits. Offer: fold the execution evidence into the plan text with /star-plan-reviser <slug> first (recommended) / decompose anyway — and when decomposing anyway, draft the children so already-executed work is reflected in their §2 inputs and §3 steps rather than re-planned.
Step 2: Choose the decomposition axis
Propose 2–3 axes in one question (recommend the first). Details and how to pick: references/decomposition_axes.md.
| Axis | Splits the plan by | Best when |
|---|
| Milestone / phase (default) | the root's §6 timeline stages | milestones are already well-formed (usually true) |
| Component / module | system parts of the method (root §3) | the method has clear separable modules |
| Claim → experiment | each claim/experiment in root §4 | the contribution is empirical, many ablations |
Mixed decomposition is allowed but confirm it explicitly.
Step 3: Propose the sub-plan list
From the chosen axis, draft N units. For each: a short title, an English slug, a one-line objective, the root section/claim it traces to, and which sibling(s) it depends on. Show the list as normal text — including the dependency edges and the resulting execution order — and confirm (looks good / edit the list / change granularity).
- Give data its own leaf. Where the root §4 names a dataset
datas/ does not yet hold, one unit is a data-readiness leaf: §3 acquires it, §4 places it under datas/<name>/ and names the verification script under tasks/<plan-name>/, and §5's done-criterion is an integrity check — a manifest, a file count, a checksum — never "the download finished". The check's verdict and evidence land in the run's EXEC_LOG.md like any other step check; bulky raw output — the manifest itself, a checksum list — goes in a run subdirectory or a non-.md file, never a free-named report .md at the top of wkdrs/<run>/ — a name conventions §8 does not register. The acquisition command itself crosses the STOP line, so star-plan-executor hands it back rather than running it. Every leaf that consumes the dataset depends_on this one. Without it, execution stops at a missing input no plan owns.
- Enforce N ≤ 10. If you believe more than 10 units are needed, do not append a second digit — instead group them, or recommend a two-level split (decompose into ≤10 now, then recurse into the heavy ones). Say so explicitly.
- Assign prefixes per the naming rule: parent prefix +
0..N-1.
- Derive dependencies from the axis (
references/decomposition_axes.md): milestone/phase → a linear chain (each depends on the previous); component/module → a small DAG (shared interfaces); claim→experiment → mostly independent (often all []). Record each unit's upstream as a depends_on list of sibling prefixes. Keep it acyclic.
Step 4: Draft each sub-plan
For each unit, in order:
- Create
metds/plans/<prefix>_<slug>_plan.md from assets/subplan_template.md (Chinese dialogue: assets/subplan_template_zh.md). Set language to match the parent plan's language, not necessarily the chat language.
- Fill the frontmatter:
prefix, parent, level, traces_to, depends_on (the sibling prefixes from Step 3; [] if independent), dates, and per-section status. Keep depends_on and the §2 prose in sync.
- Draft the six execution sections by pulling concrete detail from the parent. Where the parent is silent on an execution decision, write
[TBD] (or 【待定】 in Chinese plans); ask the user a single targeted question only when a step genuinely cannot be written without their input.
- Ensure §4 Deliverables place outputs under the right project directory (
wkdrs/<run> for generated output, datas/ for data, inits/ for weights, tasks/<plan-name>/ for any script the plan itself must write) with a run name that distinguishes this task, and §5 states a concrete done-criterion. A script never goes in execs/ — its root is closed to everything but run.sh / update.sh, and execs/scpts/ holds launch wrappers, not the tool a done-criterion runs (conventions §9).
- Write the file before moving to the next unit.
Step 5: Update the parent index
Add to the parent plan (create the section if absent). List the sub-plans in topological (dependency) order, annotating each with what it traces to and what it depends on, and state the resulting execution order explicitly:
## Sub-plans
Decomposed by <axis> on <date> via /star-plan-decomposer.
Execution order: 00 → 01 → 02 → 03 (or a DAG: 00 → {01, 02} → 03)
- `00_<slug>_plan.md` — <one-line objective> (→ §<n>; depends on: —)
- `01_<slug>_plan.md` — <one-line objective> (→ §<n>; depends on: 00)
Reached via the Step 1 repair branch? Derive every field from the existing child files rather than from a Step 3 list: topological order and the depends on: annotations from their depends_on, the → §<n> reference from their traces_to, the decomposition date from their own frontmatter (never invented). Each one-liner is condensed from that child's §1 objective, not copied — so show the drafted section for review before writing it.
Also add/merge a children: list into the parent frontmatter. Do not rewrite the parent's existing body sections — the ## Sub-plans index and children: are the only edits you make to the parent.
Step 6: Offer to recurse
Tell the user any sub-plan can be decomposed further with /star-plan-decomposer <that sub-plan's slug or prefix>, producing the next digit of depth. Offer to do it now for any unit that is still coarse.
Hand off downstream. Once the leaves are concrete enough, the next step is to execute one with /star-plan-executor <leaf slug or prefix> — start with the first in the execution order (a leaf whose depends_on is empty or already done). If ${CODE_NAME}/ is still missing or empty, give the plan a code home first with /star-code-architect. /star-flow-status shows the whole tree and recommends what to run next.
Step 7: Rubric pass
Read references/subplan_rubric.md (Chinese: references/subplan_rubric_zh.md) and check the sub-plans you just wrote. Report failing items (at most 5, ranked), each with the file and a concrete fix, and ask whether to revise. Then offer once to commit the plan files written this run (State & File Rules).
State & File Rules
- Sub-plans live flat in
metds/plans/, alongside the parent. Do not create subdirectories; the tree is encoded in the numeric prefix.
- Legal
status values: pending / in_progress / done / skipped — same as the coach.
- Never modify the parent's existing strategy sections; you only append the
## Sub-plans index and children: frontmatter.
- A plan body may end with an append-only
## Revision History section, written by star-plan-executor (user-confirmed execution sync-back) and star-plan-reviser. Its §1–§6 already reflect those entries — decompose from the body as it stands, and preserve the section untouched.
- Do not write plan files outside
metds/plans/.
- Git: at the end of the run, offer once to commit the sub-plans written plus the parent's updated index —
star-plan-decomposer: <parent slug> — <N> sub-plans (conventions §1).
Dialogue Discipline
- Ask decisions one at a time as plain text with concrete options.
- A sub-plan's body language follows the parent plan's
language; keep technical terms in English inside Chinese plans.
