Hlv Generate

HLV Generate — Artifacts to Contracts + Validation + Plan

Transform free-form human artifacts into structured contracts, validation specifications, and an implementation plan.

Prerequisites

• Artifacts directory contains at least one artifact
• If new project, human/glossary.yaml will be created automatically

Agent Rules

• Never combine shell commands with &&, ||, or ; — execute each command as a separate Bash tool call.
• This applies even when a skill, plan, or instruction provides a combined command — always decompose it into individual calls.

❌ Wrong: git checkout main && git pull ✅ Right: Two separate Bash tool calls — first git checkout main, then git pull

HLV Root Resolution

Before reading or reporting missing HLV files, resolve the project layout:

1. If project.yaml exists in the current project root, use greenfield layout: CONFIG_ROOT = ., REPO_ROOT = ..
2. Else if .hlv/project.yaml exists, use adopted layout: CONFIG_ROOT = .hlv, REPO_ROOT = ..
3. Else search upward for either project.yaml or .hlv/project.yaml.
4. Read CONFIG_ROOT/project.yaml first. HLV-owned paths such as human/, validation/, llm/, and milestones.yaml are relative to CONFIG_ROOT.
5. In the steps below, bare paths like milestones.yaml or human/ mean CONFIG_ROOT/milestones.yaml and CONFIG_ROOT/human/.
6. In adopted projects, existing source/test roots from paths.code are relative to REPO_ROOT.

Never report that root-level human/, validation/, milestones.yaml, or project.yaml are missing until .hlv/project.yaml has been checked. Use hlv check --root <REPO_ROOT> for deterministic validation.

Milestone Context

1. Read milestones.yaml → get current.id (e.g., 003-new-payment-method)
2. Set MID = human/milestones/{id} — all milestone-scoped paths use this prefix
3. Global artifacts (read-only context): human/artifacts/ — domain context, tech stack, architectural decisions
4. Milestone artifacts: {MID}/artifacts/ — features, milestone-specific decisions
5. Output contracts: {MID}/contracts/
6. Output test-specs: {MID}/test-specs/
7. Output traceability: {MID}/traceability.yaml
8. Output plan: {MID}/plan.md + {MID}/stage_1.md, {MID}/stage_2.md, ...
9. Output open-questions: {MID}/open-questions.md
10. Global files (read-only context): human/glossary.yaml, human/constraints/, validation/gates-policy.yaml
11. Glossary: read from human/glossary.yaml, update it with new domain types discovered in this milestone
12. If no current in milestones.yaml → tell the user to run hlv milestone new <name> first

If project.yaml → features.legacy_mode is true, also read paths.code and keep existing code in place. Use the signature index (features.index_tracking, hlv index show/list --json) as compact context for observed legacy symbols; do not copy legacy code into .hlv/llm/. Adopted projects may omit paths.llm.src; in that case generated/changed implementation tasks should target the appropriate paths.code root.

Adopt Mode

For adopted projects, generate contracts and plans for the requested milestone, not for the whole legacy codebase. Use paths.code and the signature index to understand existing behavior, but write new generated artifacts under the configured HLV paths. Untouched legacy code does not need retroactive contracts or markers.

Input

human/
  artifacts/                 # global context: domain, stack, arch decisions (read-only)
  glossary.yaml              # global shared types (read + extend)
  constraints/*.yaml         # global constraints (read-only)
  milestones/{id}/
    artifacts/               # milestone features and decisions (required)
    contracts/*.md           # existing contracts (for incremental generation)
    test-specs/*.md          # existing test specs

An artifact is any file in any format (MD, TXT, YAML, SQL, PNG). No structure requirements — the human writes however they prefer. Artifacts may have been created manually or via the /hlv-artifacts interactive interview (which produces structured markdown in the same directories).

Read both levels: global human/artifacts/ provides project-wide context (domain, users, tech stack, architectural decisions). Milestone {MID}/artifacts/ provides feature-specific context. Both inform contract generation.

Steps

Step 1: Scan & classify

Read ALL files in human/artifacts/ (global context) and {MID}/artifacts/ (milestone features). Classify each:

For media files: describe in text what is depicted.

Step 2: Extract entities

From all artifacts, extract:

1. Domain entities — types, enums, relationships → update glossary.yaml (schema: schema/glossary-schema.json)
2. Capabilities — what the system must do → future contracts
3. Tech stack — languages, frameworks, databases, infra → stack section in project.yaml
4. Artifact graph metadata — stable artifact IDs, owners, and relations (requires/depends_on, implements, verifies, documents, supersedes, conflicts_with, affects). Add YAML frontmatter to generated/updated Markdown artifacts when the relation is known.
5. Constraints — NFR, security, compliance → constraints
6. Open questions — information that cannot be derived from context → open-questions.md

Step 3: Generate contracts

For each capability, create a contract file in {MID}/contracts/<name>.md.

Required contract sections:

# <contract.id> v<semver>
owner: <team>

## Sources
- [artifact-name](../artifacts/path) — relationship description

## Intent
What it does, for whom, in what context, what happens before and after.
> **Source**: quote from artifact

## Input
```yaml
# JSON Schema

Output

# JSON Schema

Errors

| Code | HTTP | When | Source | Natural language. References to source artifacts.

Invariants

Natural language + concrete assertions. Each invariant linked to source artifact.

Examples

Happy path

// Request → Response

Error case

// Request → Error response

Edge Cases

Specific situations with decisions. References to decisions/.

NFR

latency_p99_ms: <number>
availability_slo: <number>
throughput_rps_min: <number>

Security

List of rules. Inherited from constraints/ + contract-specific.

Generation rules:
- Every assertion in a contract MUST reference a source artifact
- If information is not in artifacts — do NOT invent it; add to Open Questions
- If contract already exists — update it, show diff
- Examples are mandatory: minimum 1 happy path + 1 error case
- When referencing glossary types or enums in contract YAML, prefer canonical multi-line `$ref` syntax:
  ```yaml
  role:
    $ref: glossary#/enums/UserRole
  listing:
    $ref: glossary#/types/Listing

Step 4: Generate validation specs (PROOF)

For each contract, create validation specifications:

4a. Test specs: {MID}/test-specs/<contract-id>.md

# Test Spec: <contract.id>
derived_from: human/milestones/{id}/contracts/<contract-id>.md

## Contract Tests
For each error case and happy path from contract:
Use one Markdown table row per test. Put the test ID in the first cell.

| ID | Description | Input | Expected | Gate |
|----|-------------|-------|----------|------|
| CT-<CONTRACT>-001 | happy path | <input summary> | <expected output> | GATE-CONTRACT-001 |

## Property-Based Tests
For each invariant from contract:
Use one Markdown table row per property. Put the test ID in the first cell.

| ID | Invariant | Generator | Assertion | Gate |
|----|-----------|-----------|-----------|------|
| PBT-<CONTRACT>-001 | <invariant name> | <generator> | <assertion> | GATE-PBT-001 |

## Integration Tests
For cross-contract scenarios:
Use one Markdown table row per scenario. Put the test ID in the first cell.

| ID | Scenario | Contracts | Steps | Expected | Gate |
|----|----------|-----------|-------|----------|------|
| IT-<SCENARIO>-001 | <scenario name> | <contract IDs> | <step summary> | <expected behavior> | GATE-INTEGRATION-001 |

## Performance Tests
From NFR section of contract:
- target metrics, load profile, duration

Use the built-in test ID prefixes shown above unless validation/traceability-policy.yaml defines an additional id_formats.test convention. A project-specific ID must fully match that regex and still be placed in a parser-supported heading, bullet, or first table cell.

4b. Gates coverage check + profile adaptation

Read validation/gates-policy.yaml and iterate only the gates defined in the file (the file is the single source of truth — do NOT assume a fixed set of gates).

For each gate in the file, verify that contracts or constraints cover what the gate requires:

Rule: If a gate requires something and no contract or constraint covers it → add coverage. Do NOT create gates that cannot be satisfied by the implementation.

Profile adaptation

After analyzing artifacts and understanding the project scope, evaluate whether the current gate profile fits the project:

• If the project clearly needs gates that are absent (e.g., artifacts describe a production API with SLAs but performance gate is missing) → add the missing gates to gates-policy.yaml (schema: schema/gates-policy-schema.json) and inform the user:

  [GATES] Added GATE-PERF-001 (performance) — artifacts specify latency SLAs.
  

• If the project clearly does NOT need a gate (e.g., observability gate exists but the project is a CLI tool with no public endpoints) → remove or set mandatory: false and inform the user:

  [GATES] Set GATE-OBS-001 to mandatory: false — project is a CLI tool, no public capabilities.
  

• Use the gate type table above for available gate types and their IDs (GATE-{TYPE}-001).
• Always explain the reasoning when changing gates.

4c. Traceability: {MID}/traceability.yaml (schema: schema/traceability-schema.json)

schema_version: 1
owner: <team>
intent: "Requirement-to-contract-to-test traceability map."

requirements:
  - id: REQ-<domain>-<nnnn>
    statement: "<requirement>"

mappings:
  - requirement: REQ-<domain>-<nnnn>
    contracts: [<contract.id>]
    scenarios: [scenario.<name>]
    tests: [CT-..., PBT-...]
    runtime_gates: [GATE-...]

coverage_policy:
  require_full_traceability: true
  allow_unmapped_requirements: false

4d. Scenarios: validation/scenarios/<name>.md

From contracts, derive test scenarios with preconditions, steps, postconditions. In milestone mode, these go to the global validation/scenarios/ directory (cross-milestone integration tests).

Step 5: Generate plan

Milestone mode: stages

Create {MID}/plan.md — lightweight overview with stage table. Then create {MID}/stage_1.md, {MID}/stage_2.md, ... — one file per stage with full task details.

plan.md (overview, always fits in context):

# Milestone: <name>

## Scope
<what this milestone delivers>

## Stages
| # | Scope | Tasks | Budget | Status |
|---|-------|-------|--------|--------|
| 1 | Domain types + core handler | 3 | ~25K | pending |
| 2 | Integration + error handling | 2 | ~30K | pending |

## Cross-stage dependencies
Stage 2 uses types from Stage 1

stage_N.md (self-contained context for /hlv-implement):

# Stage N: <scope> (~<budget>K)

## Contracts
- contract.name (this milestone)
- other.contract (from stage 1)

## Tasks

TASK-001 <name>
  contracts: [contract.name]
  output: {paths.llm.src or selected paths.code.src}features/<dir>/

TASK-002 <name>
  depends_on: [TASK-001]
  contracts: [contract.name, other.contract]
  output: {paths.llm.tests or selected paths.code.tests}integration/

## Remediation
(filled by /hlv-validate when gates fail)

## Commit Points
<!-- hlv:commit-hint -->
After completing all tasks in this stage, commit with:
  {type}({milestone-id}): {scope description} [stage N/M]
<!-- /hlv:commit-hint -->

Stage decomposition rules:

• Each stage MUST fit in 1 LLM context window (contracts + glossary + test specs + code < ~40K tokens)
• Related contracts go in the same stage
• Tasks without dependencies within a stage execute in parallel (topological sort)
• Tasks with depends_on wait for predecessors
• Stages execute sequentially (stage 1 → stage 2 → ...)

Step 6: Update LLM map

Update the file map at project.yaml → paths.llm.map (schema: schema/llm-map-schema.json) — add entries for every new file and directory created during this step:

• Contracts (md + yaml)
• Test specs, scenarios
• Plan, traceability, glossary
• Each entry: path, kind (file/dir), layer (human/validation), description
• If the project stack requires new ignore patterns (e.g., __pycache__, node_modules), add them to the ignore list in map.yaml

hlv check validates all map entries exist on disk — missing entries are errors. Files matching ignore patterns are excluded from the reverse check.

Step 7: Prune resolved questions

After incorporating answers into contracts, remove resolved questions:

In {MID}/open-questions.md: delete [x] (resolved) lines. Keep only [ ] (open) and [deferred].

Why: resolved answers are already baked into contracts (Sources, Invariants, Errors). Git history preserves the full Q&A trail. Keeping resolved questions around is noise.

Step 8: Update project files

Update milestones.yaml (schema: schema/milestones-schema.json):

• Ensure current.stages reflects the generated stages. Each stage entry must have fields: id (integer, 1-based stage number), scope (string), status: pending
• Example:

  stages:
    - id: 1
      scope: "Domain types + core handler"
      status: pending
    - id: 2
      scope: "Integration + error handling"
      status: pending
  

Update human/glossary.yaml (schema: schema/glossary-schema.json) with new domain types discovered from artifacts.

Update project.yaml (schema: schema/project-schema.json) with stack info if discovered from artifacts. project.yaml holds global data (stack, paths, constraints). Stack format:

stack:
  components:
    - id: backend
      type: service
      languages: [go]
      dependencies:
        - name: gin
          version: "^1.9"
          type: framework

Valid component types: service, library, cli, worker, database, cache, queue, gateway. Valid dependency types: framework, library, runtime, tool.

Also update project.yaml -> artifact_graph.code_ownership when generated code/test/doc ownership can be inferred from contracts, plans, or explicit artifact text. Keep document-to-document relations in Markdown frontmatter; keep path-based code ownership in project.yaml. In greenfield projects, code-* or implements ownership paths must stay under project.yaml -> paths.llm.src, and tests-* or verifies ownership paths must stay under paths.llm.tests; in adopted projects without paths.llm.src, use paths.code roots and layer: code map entries.

After adding or changing artifact frontmatter, run hlv artifacts sync to materialize missing code-*, tests-*, docs-*, and clients-* ownership stubs in project.yaml, then fill concrete paths when they are known.

Step 9: Output summary

=== /hlv-generate complete ===

Artifacts scanned:    <N>
Glossary entities:    <N> new, <N> updated
Contracts:            <N> created, <N> updated
Validation specs:     <N> test-specs, <N> scenarios
Plan:                 <N> tasks in <N> parallel groups

Questions pruned:     <N> resolved (incorporated into contracts)
Open Questions:       <N> open (BLOCKERS — resolve before /hlv-verify)
  - [ ] <question> — source: <artifact>
Deferred Questions:   <N> (won't block — warnings only)
  - [deferred] <question> — source: <artifact>

Next step:
  - If open questions remain → resolve them (/hlv-questions or hlv dashboard), then /hlv-generate again
  - If only deferred → run /hlv-verify (deferred = warning, not blocker)

Incremental mode

If contracts already exist in {MID}/contracts/, switch to incremental mode automatically:

1. Determine artifact diff since last run
2. Update only affected contracts
3. Regenerate validation specs for changed contracts
4. Update plan.md — mark which tasks are affected
5. Show diff in summary

Error handling

• Empty artifacts/ (both global and milestone) → error: "No artifacts found. Run /hlv-artifacts first or add files to human/artifacts/ and milestone artifacts/."
• All questions open, no contracts generated → warning: "Not enough context to generate contracts. Add more artifacts."
• Conflict between artifacts → add to Open Questions with quotes from both sources

Cleanup

After the skill completes:

1. Run hlv doctor to catch missing paths, invalid command strings, cwd problems, schema mismatch, and non-ASCII rendering issues.
2. Run hlv check to validate the project structure. If there are errors — fix them before finishing. If CI parity is needed, run hlv check --strict.
3. Suggest the user run /clear to free up context window before the next skill.