--- name: majiayu000/aget-create-project source: https://app.decimal.ai/s/majiayu000-aget-create-project@1/SKILL.md source_sha256: 490fa759ae1b --- # /aget-create-project Create a new AGET project with research-informed scaffolding per PROJECT_PLAN_AGET_CREATE_PROJECT_V2.md. ## Input $ARGUMENTS ## Mode Detection | Input Pattern | Mode | Behavior | |---------------|------|----------| | Empty or blank | **Interactive** | Prompt for type and name | | ` ` | **Explicit** | Create project of specified type | | `` only | **Inference** | Infer type from topic and AGET domain | | Input contains evidence statements (interview data, gap analysis, L-doc citations, cross-reference synthesis) | **Evidence-rich** | Abbreviated research phase — use provided evidence, skip redundant searches | ### Evidence-Rich Mode **Detection**: Input contains 2+ of: specific L-doc references (L###), quantitative data (percentages, counts, session numbers), named sources (supervisor interviews, audit results), or cross-reference synthesis ("both supervisors agree..."). **Behavior**: Research Phase (Step 3) still executes per C-CP-003, but in abbreviated form: - Step 3.1 (identity): **Runs** — always needed for project context - Step 3.2 (L-docs): **Abbreviated** — search only L-docs explicitly cited in input, don't broad-search by topic keywords - Step 3.3 (sessions): **Abbreviated** — note current session only (evidence was gathered in-session) - Step 3.4 (similar projects): **Runs** — conflict check always needed - Step 3.5 (vocabulary): **Runs** — vocabulary grounding always needed **Rationale**: When the user provides rich evidence context (e.g., from supervisor interviews or gap analysis conducted in the same session), broad keyword searches duplicate work already done. The abbreviated research phase respects C-CP-003 (research is NOT skipped) while avoiding redundancy. ## Supported Project Types | Type | Template | Output Location | |------|----------|-----------------| | `research` | `templates/poc/RESEARCH_PROJECT_PLAN.template.md` | `planning/PROJECT_PLAN_.md` | | `poc` | `templates/poc/PROPOSAL.template.md` | `planning/poc-proposals/POC-XXX-.md` | ## Project Creation Process ### Step 0: Scope-Fit Validation (#295) **Purpose**: Verify the proposed project falls within this agent's domain and scope boundaries before investing in research or scaffolding. 1. Read `.aget/identity.json` — extract `domain`, `north_star`, `archetype` 2. Read `governance/SCOPE_BOUNDARIES.md` — extract in-scope and out-of-scope areas 3. Evaluate alignment between the project topic and the agent's domain/scope 4. If **misaligned**: WARN with explanation and allow user override: ``` WARN: Scope-fit concern. Project topic: "" Agent domain: "" Concern: Override? The principal may proceed if this is intentional. ``` 5. If **aligned**: proceed silently (no output needed) **Rationale**: L616 — projects created outside an agent's domain waste effort and produce artifacts that don't integrate with the KB. Scope-fit validation is the cheapest possible check (2 file reads) with the highest leverage. ### Step 1: Input Analysis Parse $ARGUMENTS to determine: - Project type (research, poc, or infer from topic) - Project name - If empty in interactive mode, prompt user **Headless mode**: If no arguments and headless, return error: ``` Error: Project type and name required in headless mode. Usage: /aget-create-project Types: research, poc ``` ### Step 2: Conflict Check Check for existing project: ```bash ls planning/ | grep -qi "" ``` If project exists, WARN and ask for confirmation before proceeding. ### Step 3: Research Phase (KEY DIFFERENTIATOR) Before scaffolding, gather AGET context: #### 3.1 Read AGET Identity ```bash cat .aget/identity.json ``` Extract: - `north_star`: Agent's purpose - `domain`: Research focus areas - `archetype`: Agent type #### 3.2 Search Relevant L-docs ```bash grep -l "" .aget/evolution/*.md ``` For each found L-doc, extract: - L-doc ID and title - Key lesson - Application to this project #### 3.3 Search Prior Sessions ```bash grep -l "" sessions/*.md ``` Note related sessions for Traceability. #### 3.4 Search Similar Projects ```bash grep -l "" planning/PROJECT_PLAN_*.md ``` If similar project exists, WARN about potential duplication. #### 3.5 Extract Vocabulary Terms ```bash grep -B2 -A5 "" specs/CLI_VOCABULARY.md ``` Identify vocabulary terms for Traceability section. ### Step 3.6: Spec-First Conformance Baseline (L616, L617, #313) **Purpose**: If the project targets a spec-governed artifact, read the governing spec and create a conformance baseline BEFORE designing gates. This enforces deductive (spec conformance) before inductive (peer comparison) planning per MP-1. 1. Determine if the project will create or modify a spec-governed artifact 2. If **spec-governed**: a. Read the governing spec (e.g., `aget/specs/AGET_PROJECT_PLAN_SPEC.md`) b. Extract all SHALL/SHOULD requirements relevant to the target artifact c. Create a conformance matrix: requirement ID → current status (met/unmet/partial) d. Include the conformance matrix in Phase -1 of the scaffolded plan e. Design gates to address unmet requirements in priority order (SHALL before SHOULD) 3. If **greenfield** (no governing spec exists): a. Note "No governing spec identified — greenfield project" in Phase -1 b. Proceed to Step 3.7 **Rationale**: L617 — inductive planning (looking at peer projects for gate structure) bypasses spec conformance. The spec defines what the artifact MUST do; gates should be derived from spec requirements, not from what similar projects did. ### Step 3.7: ADR-008 Precondition Check (#313) **Purpose**: Verify the ADR-008 prerequisite chain is satisfied before creating a plan that assumes infrastructure exists. **ADR-008 ordering**: L-doc evidence → SOP → Spec → Skill/Tool (Generator) 1. Determine the artifact type the project will produce (SOP, spec, skill, template, etc.) 2. Check prerequisites for that artifact type: - **Skill**: requires governing spec + SOP to exist - **Spec**: requires SOP or L-doc evidence to exist - **SOP**: requires L-doc evidence (2+ executions or 3+ L-docs per L436) - **L-doc**: no prerequisites (generates evidence) - **Research project**: exempt (generates the evidence other types require) 3. If **prerequisites missing**: WARN with specific gaps: ``` WARN: ADR-008 precondition gap. Target artifact: Missing prerequisite(s): - - Consider: Create prerequisites first, or note gap in project plan. ``` 4. If **prerequisites met**: proceed silently **Rationale**: L624 — skills modified without SOPs or specs create governance gaps. ADR-008 defines the progression; this step enforces it structurally. ### Step 3.8: Artifact Type Fitness Assessment (#315) **Purpose**: Cross-reference the proposed artifact type with the domain's ADR-008 maturity level to assess fitness. 1. Identify the proposed artifact type and its domain 2. Assess domain maturity: - **Mature domain** (3+ L-docs, SOP exists, spec exists): any artifact type appropriate - **Emerging domain** (1-2 L-docs, no SOP/spec): recommend research project or L-doc first - **Novel domain** (0 L-docs): recommend research project only 3. If **artifact type exceeds domain maturity**: WARN with recommendation: ``` WARN: Artifact type may exceed domain maturity. Proposed: (requires ) Domain maturity: ( L-docs, , ) Recommendation: Consider first to build evidence base. ``` 4. If **appropriate**: proceed silently **Rationale**: L617, ADR-008 — creating a spec for a domain with no L-doc evidence produces speculative requirements. Maturity assessment prevents premature abstraction (L103). ### Step 4: Template Selection and Scope Estimation (#391) **4a: Scope Estimation** (before template routing): Estimate project scope based on deliverables, complexity, and evidence from research phase: - **Small** (<1 day): Single artifact, few gates, well-understood domain - **Medium** (<2 weeks): Multiple artifacts, 3-5 gates, some research needed - **Large** (2+ weeks): Cross-cutting changes, 5+ gates, significant research If scope suggests a **smaller template** would be more appropriate (e.g., a 1-day task routed to RESEARCH_PROJECT_PLAN): ``` WARN: Scope-template mismatch. Estimated scope: Selected template: