AGENT SKILL

plugin-conception

Turns a rough idea, imported draft, or recent-session evidence of friction, failure, or opportunity into a decision-ready classification (Create, Enhance, Repair, Consolidate, Reposition, Retain, Reject/Defer) before plugin-ideation's interview or plugin-planning/Fix begins. Runs a shared normalize/classify/shallow-overlap-check core from either starting point, then produces a Conception Brief whose depth matches the classification — light for Create (hands off to plugin-ideation unchanged), full for every other outcome (marketplace integration, implementation plan, baseline contract). Use when the user asks "should this be a new plugin or an enhancement", "classify this idea", "is this new or should it extend X", "turn these session findings into a concept", or has a rough idea or session evidence that needs qualifying before ideation or planning/Fix runs. Not a replacement for plugin-ideation's interview or plugin-planning's component inventory — this is the qualifier that precedes both.

License 许可
Apache-2.0
最近更新
2026-11-15
Stars
★ 0
Forks
0

01 前言 Front Matter

name
plugin-conception
description
Turns a rough idea, imported draft, or recent-session evidence of friction, failure, or opportunity into a decision-ready classification (Create, Enhance, Repair, Consolidate, Reposition, Retain, Reject/Defer) before plugin-ideation's interview or plugin-planning/Fix begins. Runs a shared normalize/classify/shallow-overlap-check core from either starting point, then produces a Conception Brief whose depth matches the classification — light for Create (hands off to plugin-ideation unchanged), full for every other outcome (marketplace integration, implementation plan, baseline contract). Use when the user asks "should this be a new plugin or an enhancement", "classify this idea", "is this new or should it extend X", "turn these session findings into a concept", or has a rough idea or session evidence that needs qualifying before ideation or planning/Fix runs. Not a replacement for plugin-ideation's interview or plugin-planning's component inventory — this is the qualifier that precedes both.
allowed-tools
Read Glob Grep AskUserQuestion Write Bash(date:*) Skill
argument-hint
[rough idea, problem statement, or session-evidence source]

02 资源文件 3

03 SKILL.md 正文

Plugin Conception

Turns either a rough idea or evidence of an unmet need into a decision-ready classification before plugin-ideation's interview, plugin-planning's component inventory, or plugin-lifecycle-maintenance's Fix phase begins. This is a pre-ideation, pre-planning qualifier, not a replacement for the deeper work either downstream skill already does — it establishes that a proposed change is useful, distinct, appropriately placed, bounded, and worth the next phase's cost, then hands off.

Conception is a classification pass, not an interview. Unlike plugin-ideation's multi-round back-and-forth, this skill interviews the requester only where a missing answer would change the classification or the shallow overlap read — never to plugin-ideation's depth, and never re-asking questions plugin-ideation will ask again immediately after a Create handoff.

Quick Start

  1. Route — from-scratch idea, or recent-session evidence? (Entry Routes A/B)
  2. Normalize and classify — build a problem frame, label assumptions, then classify: Create / Enhance / Repair / Consolidate / Reposition / Retain / Reject-Defer (Steps 1-2)
  3. Check overlap, shallow only — repository-metadata depth; confirms or revises the classification (Step 3)
  4. Define the concept, non-Create only — full depth, implementation plan, workflow if procedural (Steps 4-6; Create skips straight to Step 7)
  5. Decide and hand off — approve / revise / merge / defer / reject; write the Conception Brief (Step 7)

When to Use

  • Before plugin-ideation's interview, to confirm a rough idea is actually worth that interview (or is really an Enhance/Consolidate in disguise)
  • Before plugin-planning/plugin-lifecycle-maintenance's Fix phase, to turn session evidence, a finding, or an enhancement request into a bounded, evidence-backed concept with a baseline contract
  • Invoked automatically as Phase 1 (Conceive) of plugin-lifecycle-upstream, or as a step inside plugin-lifecycle-maintenance's improve-a-plugin/enhance-a-plugin workflows, immediately after their own human finding-selection gate

When NOT to Use

  • The concept is already a Create classification with a clear problem statement and nothing left to qualify — go straight to plugin-ideation; running this skill first only to immediately hand off is needless ceremony when the classification was never in doubt
  • A narrow repair with an already-known fix and an already-accepted finding — see "Bypass for narrow repairs" under Step 2; hand off directly to Fix instead
  • Deep per-component overlap search, activation-trigger overlap analysis, or naming-candidate validation — that's plugin-ideation Step 3-4's job (Create) or the type-matched reviewer agents' job (everything else), never this skill's
  • Scaffolding components, editing production plugins, or running QA fixes — this skill produces a brief, never files
  • Already committed to running the full guided pipeline (Conceive→Ideate→Plan→Design→Build), not just the classification step — use plugin-lifecycle-upstream instead; it dispatches here automatically as Phase 1, so invoking this skill directly is for when the classification decision itself is the goal
  • Already committed to acting on retro/comparison findings against an already-shipped plugin, with apply/test/commit follow-through — use plugin-lifecycle-maintenance's improve-a-plugin/ enhance-a-plugin workflows instead; they dispatch here automatically after the human finding-selection pick, so invoking this skill directly is for classifying a rough idea or evidence source before any pipeline commitment exists

Entry Route A: Start From Scratch

Use when the starting point is a rough idea, problem statement, desired capability, or proposed plugin/skill. If $ARGUMENTS is given, treat it as the rough idea or problem statement directly — the same way plugin-ideation's own Step 2 treats a detailed $ARGUMENTS — and reflect it back rather than re-asking for it from scratch. Minimum intake, deliberately light since a Create outcome hands off to plugin-ideation's own deeper interview rather than duplicating it here:

  • the problem or opportunity;
  • the concrete user and usage situation, at whatever resolution the requester already has it;
  • the intended outcome;
  • known constraints and explicit non-goals, as far as they're already known;
  • any existing component the requester believes is adjacent.

Interview the requester only where a missing answer changes classification or the shallow overlap read. Do not fabricate evidence that does not exist, and do not require a retrospective the requester never asked for.

Entry Route B: Start From Recent-Session Evidence

Use when repeated friction, failures, gaps, or opportunities have been observed in one or more recent sessions — session-analysis reports, build handoff reports, validation/grading/comparison/security/ completeness/consistency findings, user corrections, or durable planning artifacts. If $ARGUMENTS names one of these sources directly (a report path, or one entry from a findings bundle or approved suggestion list an upstream workflow's own human-decision gate already produced), treat it as the evidence source for Step 1 below rather than re-deriving it. This skill classifies and briefs exactly one candidate per invocation — when a caller's own gate approved several candidates at once (e.g. plugin-lifecycle-maintenance's multiSelect finding-selection pick), it invokes this skill once per candidate, never once with the whole list; this skill's own Step 1 never fans a single invocation out into multiple classifications or multiple briefs. Session evidence is a seed, not the scope of the concept. See references/evidence-routing.md for the full 6-step evidence-handling procedure (identify source, recheck currency, separate symptoms from underlying need, merge duplicates, discard stale/non-actionable, obtain explicit approval before promoting to a planned change) and the evidence-source list. If evidence is insufficient, fall back to Entry Route A's focused interview — never require a full retrospective the desired concept doesn't actually depend on.

Most session evidence classifies as Enhance, Repair, or Consolidate rather than Create — those classifications get the full Conception Brief (Steps 4-5 below), since plugin-lifecycle-maintenance's existing workflows have no equivalent deep-definition step of their own.

Step 1: Normalize Intent and Evidence

Build a concise problem frame from whichever entry route ran:

  • problem or opportunity;
  • target user and concrete scenario;
  • desired outcome;
  • supporting evidence or assumptions;
  • constraints and non-goals;
  • success signals.

Label every assumption explicitly. Every piece of evidence must identify its source. Never rely on an unverified claim that a capability is absent or broken — check the current repository state instead of trusting a stale report.

Step 2: Classify the Proposed Change

Classification Meaning Likely destination
Create No suitable existing capability; add a new plugin or component plugin-ideation, then the rest of the upstream pipeline
Enhance Preserve an existing component and add or improve behavior plugin-planning (if new components are needed) or directly to Fix
Repair Restore behavior that contradicts the current contract Directly to Fix — usually skips the full brief (see below)
Consolidate Multiple components overlap and should be merged or have clearer boundaries plugin-planning plus consistency/activation review
Reposition Capability exists but belongs in a different plugin or functional group plugin-planning and marketplace restructuring
Retain Current behavior is adequate; no implementation is justified Stop with rationale
Reject/Defer Benefit, evidence, feasibility, or priority is insufficient Stop or record a revisit condition

The classification may change after Step 3's overlap check — an apparent Create may turn out to be Enhance or Consolidate, and, just as often, an apparent Enhance may turn out to be a genuine Create once the shallow check shows nothing adjacent actually exists.

Bypass for narrow repairs: a Repair classification with an already-known, narrowly-scoped fix and an already-accepted finding does not need the full brief (Steps 4-5) — record the classification and evidence, then hand off directly to Fix. This keeps the skill proportional rather than adding ceremony to a one-line fix.

Step 3: Check Overlap and Marketplace Fit — Shallow, Classification-Level Only

This check is deliberately shallow. Its only job is to confirm or revise Step 2's classification — never the exhaustive per-component search plugin-ideation runs immediately after a Create handoff, and never the deep evidence-gathering the finding-source tools already ran before this skill started.

Inspect, at repository-metadata depth only:

  • plugin manifests and plugin purposes;
  • skill, agent, command, hook, and rule descriptions;
  • existing lifecycle ownership;
  • related draft concepts and generated reports.
Overlap Meaning
None Step 2's classification likely holds; proceed
Partial Classification may need to shift toward Enhance/Consolidate; reconsider before proceeding
Full Classification shifts to Retain or Enhance; a Create outcome is very unlikely to survive this finding

Do not, here: activation-trigger overlap analysis, cross-component reference tracing, or naming-candidate validation — escalating past metadata-level inspection is the single most common way this skill ends up duplicating plugin-ideation. Marketplace placement follows the repository convention: a plugin uses <domain>-kit or <domain>-devkit, exactly one hyphen immediately before the suffix. For Create outcomes, this skill never proposes a name — that's plugin-ideation Step 4's job. For every other outcome, record the existing plugin's placement, not a new name.

If the classification settled here is Retain, Reject, or Defer: stop now, per Stop Conditions below — do not proceed to Steps 4-7, and do not write a Conception Brief file. State the classification and rationale directly in the response instead. This is the only path that produces "no brief written" — see the Testing & Validation section's Retain scenario.

Step 4: Define the Concept — Full Depth Only for Non-Create

Create: skip this step entirely. Step 1's problem frame plus the classification and Step 3's shallow overlap result is the whole Create handoff payload.

Enhance / Repair (full-brief) / Consolidate / Reposition: the accepted concept must specify a one-sentence value proposition, target users and triggering situations, in-scope capabilities, explicit non-goals, whether the target is a new component or a change to an existing one (never a new plugin — that would be a Create by definition), the proposed marketplace owner and neighboring components, dependencies/consumers/lifecycle handoffs, compatibility requirements for existing behavior, measurable acceptance criteria, principal risks and mitigations, and a baseline contract (behavior that must remain unchanged, observed deficiency and evidence, proposed delta, affected callers/docs/tests/activation triggers, rollback or rejection conditions).

Step 5: Shape the Implementation Plan

Create: skip this step — plugin-planning (reached after plugin-ideation completes) owns component-type/count/depth decisions.

Every other non-bypassed classification: define target files/component directories, required new or modified components, dependency/integration changes, tests or evaluations that demonstrate the concept, documentation changes, validation/review/commit boundaries, and a recommended execution order. Detailed prompts, scripts, schemas, and line-level edits still belong in the subsequent planning/design phase — not here.

Step 6: Describe the Workflow When the Concept Is Procedural

Include a workflow only when the proposed capability coordinates three or more dependent states, gates, or handoffs — entry conditions, ordered phases, artifacts consumed/produced, human approval gates, stop/ retry/resume behavior, downstream owner per handoff, completion criteria. A single focused skill does not need an invented workflow merely to fill this section — mark it not applicable with a one-sentence reason instead.

Step 7: Decide and Hand Off

Entry: the classification is Create, Enhance, Repair (full-brief path), Consolidate, or Reposition — a Retain/Reject/Defer classification already stopped before reaching this step (see the gate at the end of Step 3).

Present the completed brief via AskUserQuestion: approve and proceed / revise / merge with an existing concept / defer with a revisit condition / reject or retain the current state instead. This last pair of options is the user overriding an already-written brief post-hoc (e.g. approving the classification but deciding on reflection the change isn't worth doing) — a different case from the classification-level Retain/Reject/Defer stop above, which never reaches this step at all. Only an approved concept proceeds — this skill never auto-applies findings or silently turns evidence into a feature.

  1. Get a timestamp: date -u +%Y-%m-%dT%H-%M-%SZ
  2. Write to .claude/output/plugin-conception/<concept-slug>-<timestamp>.md per references/conception-brief-template.md — for Create, the light variant its own blockquote instructs (the section list lives there, not restated here, to avoid drift); for Enhance/Repair (full-brief)/Consolidate/Reposition, all 12 sections in full
  3. Present the artifact link first, then the summary:
📄 Conception Brief written: `.claude/output/plugin-conception/<concept-slug>-<timestamp>.md`

Hand-off target follows the classification:

  • Create → plugin-ideation, with the light brief as its input.
  • Enhance / Consolidate / Reposition → plugin-planning if new or restructured components are implied, otherwise directly to plugin-lifecycle-downstream's Phase 8 (Consolidated Fix).
  • Repair (full-brief path) → directly to plugin-lifecycle-downstream's Phase 8 (Consolidated Fix).
  • Retain / Reject / Defer → stop; no downstream hand-off.

Phase 8 needs a findings bundle, not a brief — this skill never invokes it directly. plugin-lifecycle-downstream's Phase 8 external-entry contract accepts only a schema-validated findings bundle (plugin-rulebook/references/evidence-schema.md's Finding/Report-Revision shape, with a baseline_commit and scope manifest) and refuses anything else — a Conception Brief path does not satisfy it, and this skill has no Bash(git log:*) grant to produce baseline_commit even if it tried. Where Phase 8 is named as the destination above, the reformatting-and-invoking step belongs to whichever orchestrator called this skill (e.g. plugin-lifecycle-maintenance's improve-a-plugin/enhance-a-plugin Step 4 already does this conversion) — never to this skill itself.

Standalone invocation: for a plugin-ideation/plugin-planning hand-off, ask via AskUserQuestion before invoking the target — never invoke it silently. For a Phase 8 destination, there is no orchestrator present to do the conversion: state the classification and that a findings-bundle conversion is needed before Phase 8 can be reached, and stop — never attempt to invoke plugin-lifecycle-downstream directly.

Nested invocation: when this skill runs as a step inside another orchestrator — plugin-lifecycle-upstream's Phase 1 (Conceive), or plugin-lifecycle-maintenance's improve-a-plugin/enhance-a-plugin Conceive step — the brief-content approval above (approve/revise/merge/defer/reject) still runs, but the hand-off invocation does not: this step ends at "write the brief and report its classification." The calling orchestrator owns deciding where each classified candidate goes next, since it may be routing several candidates from one run — that decision may be a single consolidated AskUserQuestion (as plugin-lifecycle-upstream's Gate 1 does) or a deterministic routing rule stated plainly to the user with no further ask (as plugin-lifecycle-maintenance's improve-a-plugin/enhance-a-plugin Step 4 does, routing by classification alone). Either way, never ask twice for the same decision — see those two callers for how each owns this instead.

Resume Behavior

  • A draft brief resumes at the first incomplete section (light briefs have fewer sections to resume through).
  • An approved brief resumes at handoff after the user confirms it is still current.
  • Evidence is rechecked when the underlying target has changed since approval.
  • A legacy Concept Card may still be read directly by plugin-ideation without ever passing through this skill — conception is not a mandatory gate in front of an already-existing Concept Card.

Stop Conditions

Stop without implementation when: full overlap makes a new component unnecessary; current behavior satisfies the verified need; evidence is stale or already resolved; the concept lacks a concrete user outcome; marketplace ownership cannot be decided without a user choice; risks or cost exceed the accepted benefit; or the user defers or rejects the concept. A clean stop is a valid result and must include its rationale — never silently drop a concept without stating why.

Testing & Validation

Eval evidence: evals/plugin-conception/evals.json — 8 scenarios (Quick Workflow, workspace/iteration-1), 7/8 eval-covered (scenario 1's eval is invalidated by a scripted-premise flaw unrelated to this skill's behavior — see its own grading.json); the remaining scenarios below are design-review-verified.

  1. From-scratch Create — a rough idea with no session evidence; confirm the light brief (per references/conception-brief-template.md's own light-variant section list) is produced and handed to plugin-ideation, never a padded-out full brief
  2. Session-evidence Enhance — evidence pointing at friction with an existing component; confirm the full brief is produced (all 12 sections) with a populated baseline contract
  3. Full overlap — an idea that duplicates an existing component; confirm the classification shifts to Retain/Enhance and the skill stops with rationale rather than proceeding toward Create
  4. Stale evidence — session evidence describing a since-fixed gap; confirm Step 1's current-state reverification catches this and the evidence is discarded, not promoted
  5. Narrow-repair bypass — a Repair classification with an already-known fix and an accepted finding; confirm Steps 4-5 are skipped and the skill hands off directly to Fix
  6. Consolidate — multiple overlapping components found during Step 3; confirm the full brief is produced and plugin-planning plus consistency/activation review are named as the destination
  7. Retain/no-work outcome — evidence that current behavior is already adequate; confirm a clean stop with stated rationale, no brief written
  8. Create brief stays light — construct a Create classification with rich session evidence available; confirm the written brief still marks Scope and Boundaries, Existing-Component Baseline, Implementation Plan, and the "eventual implementation success" half of Acceptance Criteria as not applicable, rather than restating what plugin-ideation/plugin-planning will produce

Quality gates:

  • [ ] Step 3's overlap check never exceeds repository-metadata depth — no activation-trigger analysis, cross-component reference tracing, or naming-candidate validation
  • [ ] A Create classification's brief always marks Scope and Boundaries, Existing-Component Baseline, Implementation Plan, and the "eventual implementation success" half of Acceptance Criteria not applicable — never a padded-out 12-section document
  • [ ] The narrow-repair bypass (Step 2) is only ever taken when the fix is already known and the finding already accepted — never for an undefined or contested repair
  • [ ] Entry Route B evidence is always rechecked against current repository state before promotion — never copied forward from a source artifact unverified
  • [ ] Step 7's decision is always made via AskUserQuestion — this skill never auto-applies a finding or silently promotes evidence into a planned change
  • [ ] Standalone invocation: the hand-off target is always invoked via AskUserQuestion confirmation first — never invoked silently. Nested invocation (inside plugin-lifecycle-upstream's Phase 1 or plugin-lifecycle-maintenance's Conceive step): no hand-off is invoked from here at all — only the brief-content approval fires; the calling orchestrator owns the hand-off routing decision (a consolidated ask, or a plainly-stated deterministic rule)
  • [ ] The Conception Brief path is always under .claude/output/plugin-conception/
  • [ ] A clean stop (Retain/Reject/Defer/full overlap/stale evidence) always states its rationale — never a silent drop
  • [ ] A classification-level Retain/Reject/Defer outcome always stops at the end of Step 3, before Steps 4-7 run — no Conception Brief file is ever written for it

Reference Guide

Resource Type Purpose
scripts/smoke_test.py In this skill This skill's own persisted smoke test (frontmatter validity, referenced-file existence, Bash-scope grant consistency) — re-run after any SKILL.md edit
evals/plugin-conception/ Repo root Persisted skill-tester Quick Workflow eval suite (8 scenarios, 7/8 covered)
references/conception-brief-template.md In this skill Light and full Conception Brief templates used in Step 7
references/evidence-routing.md In this skill Entry Route B's evidence sources and the 6-step evidence-handling procedure
plugin-ideation skill Sibling skill Create-classification hand-off target — owns the interview, deep overlap search, and naming
plugin-planning skill Sibling skill Hand-off target for Enhance/Consolidate/Reposition outcomes implying new or restructured components
plugin-lifecycle-downstream skill Sibling skill Owns Phase 8 (Consolidated Fix) — hand-off target for Repair and no-new-component Enhance outcomes
plugin-lifecycle-upstream skill Sibling skill Runs this skill as Phase 1 (Conceive)
plugin-lifecycle-maintenance skill Sibling skill Runs this skill as a step inside improve-a-plugin/enhance-a-plugin, after the human finding-selection gate

04 同仓库更多 Skills

AndreHahm
AndreHahm/andres-cc-marketplace

demo

Fixture skill used by scripts.marketplace_ci sync tests.

★ 0
AndreHahm
AndreHahm/andres-cc-marketplace

export-demo

Fixture skill used by scripts.marketplace_ci export tests.

★ 0
AndreHahm
AndreHahm/andres-cc-marketplace

work-to-development

Validate an accepted Linear Issue's readiness for implementation (criteria, dependencies, duplicate/existing artifacts) and request governed branch/worktree creation through git-kit:starting-work — recording work-started Git/GitHub evidence and a deliberate Linear Started transition only after Git identity is read back and verified. Use when asked to start work on a Linear issue, begin development for an accepted issue, or prepare an accepted issue for implementation. Never creates a branch directly — always delegates to git-kit.

★ 0
AndreHahm
AndreHahm/andres-cc-marketplace

work-linking

Create and maintain stable cross-system links and authority labels between Notion and Linear records, and detect/repair drift between them. Use when asked to link a Notion record to a Linear record, check for drift between Notion and Linear, or repair a broken/stale link. Repairs only the bounded non-authoritative copy or link — never chooses by newest timestamp, and never creates a reverse-write loop against either system's own authority.

★ 0
AndreHahm
AndreHahm/andres-cc-marketplace

status-and-learning

Produce a dated, explicitly non-live Linear-to-Notion progress summary, or capture an outcome/learning after work completes. Use when asked to summarize progress to Notion, report portfolio status, or capture what was learned from a completed piece of work — not to disposition open follow-up items (see open-item-management for that). Never runs automatically or continuously — every summary is a deliberate, approved snapshot, not a live sync.

★ 0
AndreHahm
AndreHahm/andres-cc-marketplace

repository-gates

Discover, delegate, record, and invalidate this repository's actual Git/GitHub lifecycle gates (pre-commit, pre-push, PR required checks, Codex delta review, review requirements, merge rights) and the repository-policy provider profile that maps governed operations to git-kit. Use when asked what gates apply to this repository, to discover repository policy before starting Wave 2 work, or to check whether prior gate evidence is still valid after new commits. Discovery and delegation only — never executes a gate's own internal logic, and never invents a gate the repository doesn't actually define (in particular, never assumes a "Review Changes" gate exists).

★ 0