Skip to content

ayoubben18/ab-method / .agents/skills/critique-plan

critique-plan

Verified

Pre-implementation critic. Spins up a read-only domain critic that challenges a drafted plan — a task's missions or a roadmap's task graph — against the ubiquitous language, CONTEXT.md, ADRs and the documented seams, flags plans that bend a feature around a design it should reshape, and pushes back ONLY on genuine conflicts. Silent when the plan is sound. Use before implementing (from create-task / create-roadmap) or standalone to stress-test a plan against the project's domain model.

Critique Plan (pre-implementation)

Stress-test a drafted plan against the project's domain model before any code is written. This skill produces pushbacks, not suggestions — it never gold-plates a plan that is already sound.

Silence is the expected outcome. A plan that speaks the canonical language and respects the documented decisions gets a one-line "No objections." Do not manufacture concerns to look thorough.

ALWAYS check .ab-method/structure/index.yaml FIRST for where the domain model lives — paths are user-configurable.

Look for it under the project root (the current working directory) first — a project's own copy is how it customises its paths, so it always wins. Only if the project has none (AB Method installed as a plugin rather than with npx ab-method), read the bundled default: ../../../.ab-method/structure/index.yaml relative to this SKILL.md. Either way, every path the index names is relative to the project root, never to the folder the bundled file lives in.

"The plan" is the drafted mission list together with the tracker's ## Placement and ## Adopt or Build sections (from create-task), the task DAG (from create-roadmap), or whatever the user pastes in (standalone).

Process

1. Load the domain model

Read only what exists; skip missing files silently (don't flag them or offer to create them): UBIQUITOUS_LANGUAGE.md, CONTEXT.md (or CONTEXT-MAP.md + each src/<context>/CONTEXT.md), docs/adr/, docs/architecture/tech-stack.md, each layer's docs/architecture/<layer>/index.md (open a topic file only when the plan touches it), and docs/seams/index.md (open a seam doc only when the plan touches that seam).

2. Spin up ONE read-only domain critic (subagent)

Spawn a single subagent — domain-critic — with the plan verbatim, the files from Step 1, and the rule that it is read-only: it returns pushbacks as text and edits nothing. Isolating it keeps the critique out of the planning context. Its brief:

Fire ONLY on a genuine conflict, one of:

  • Terminology drift — the plan names a concept differently from the glossary/CONTEXT.md, or reuses a canonical term for a new meaning.
  • Wrong bounded context — work placed in the wrong context, or a unit that straddles a documented boundary.
  • Contradicts an ADR — reverses a recorded decision and the friction is real enough to reopen it. Cite ADR-NNNN.
  • Bad graph seam (roadmap) — a depends-on edge crosses a seam the wrong way, two "independent" tasks share a domain concept, or a task is mis-scoped (an epic, or a single mission dressed as a task).
  • Reinvents a named concept — introduces a new abstraction for something the domain model names.
  • Bypasses a documented seam — a mission does work a seam in docs/seams/ owns without going through it (direct fetch beside the data-fetching seam), or changes a seam's interface without an introduce placement or a pass through the seam design gate (create-task § 3). Cite the seam doc and its Callers never line.
  • Fits around a design it should reshape — a mission adds a special case, flag, adapter or second path to get past the current design, cuts or stages a capability the task's ## Outcome asks for, or defers part of it to "later" with no real boundary forcing the split, while the tracker's ## Architecture says fit or is missing (create-task § 3a). Cite the mission and the workaround, and sketch the reshape that removes it. Effort is not a reason to keep the fit; a real constraint (an ADR, an external contract, persisted data) is, so don't fire when ## Architecture names one.
  • Unvetted dependency — a mission installs or newly imports a package (including one only in the lockfile) with no adopt, covered or adopt (override: …) row for it in ## Adopt or Build, a mission that installs a different version than the row checked, or a row checked more than 30 days ago. This and the reshape check are the two not anchored in the domain model. Cite the mission and the package.

For each, return What (the mission/task/decision), Conflicts with (the exact term / CONTEXT.md section / ADR-NNNN / docs/seams/<slug>.md), Why it matters (concrete cost, not taste), Suggested resolution.

Example pushback: "Mission 3 calls it archiveOrder, but the glossary defines archiving as retention only — this mission also stops billing, which is Cancellation. Rename to cancelOrder so the code matches the domain, or the two concepts will blur across the codebase."

A parked question is not a gap. If the plan carries ⚠️ UQ-n markers, read the task's unresolved-questions.md and pass it to the critic as context. A recorded black box is a decision the user made — to ship a placeholder rather than guess — so the critic must not fire on "this mission is underspecified" or "the answer to UQ-1 should be X". It fires only if the placeholder itself conflicts with the domain model: it contradicts an ADR, quietly redefines a canonical term, or the parked question turns out to be a terminology question wearing a behaviour costume (those aren't parkable — they belong in CONTEXT.md). A marker with no matching entry is worth one line: the plan claims a black box that isn't recorded anywhere.

Out of scope for this critic: implementation quality, performance, tests, code style, "you could also…" ideas — anything not anchored in the domain model. Those belong to the post-implementation review-implementation skill. With nothing anchored, the critic returns exactly: No objections — the plan is consistent with the domain model.

Every return, clean or not, ends with one evidence line: Checked: <the seam docs, ADRs and rules it actually read, and the files it opened>. A clean result with no evidence line is a pass that didn't happen: re-run the critic.

3. Surface pushbacks — advisory, never blocking

Bring the pushbacks back into the planning session. The user resolves each their way:

  • Accept → amend the plan (rename, re-scope, fix the edge, move contexts) right there.
  • Dismiss → drop it. If the dismissal rests on a load-bearing reason a future planner would need in order not to re-raise it, offer to record an ADR (../domain-model/ADR-FORMAT.md). Skip ephemeral ("not now") and self-evident reasons.

When a resolution sharpens a term, update CONTEXT.md inline (../domain-model/CONTEXT-FORMAT.md) — same discipline as /domain-model.

If the critic returned "No objections," say so in one line and move on. Don't pad it.

/domain-model is a full interactive re-grill of the design; critique-plan is a single-pass, silent-by-default gate — one critic, real conflicts only, then straight back to the workflow.

finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the workStars: 319kUpdated Claude CodeCodexVerifiedInstall
systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixesStars: 319kUpdated Claude CodeCodexVerifiedInstall
test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation codeStars: 319kUpdated Claude CodeCodexVerifiedInstall
skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.Stars: 196kUpdated Claude CodeCodexVerifiedInstall