v3.1.10 — Spec Guardrails 3.1.x

Spec Guardrails 3.1.10 — Process + Brakes modes, Copilot/Codex adapters, validate-traceability, validate-quick. npm: @luizsantiago/spec-guardrails

v2.1.0 — loop-plan parallel waves

Highlights

• loop-plan — analyzes tasks.md and shows the next Execute wave with parallel groups (disjoint Files)
• /loop orchestration — each round runs loop-plan, dispatches sub-agents for parallel groups, or works inline for single tasks
• Docs reframed: parallel execution was already in task-graph-engineering.md + sub-agents.md; now first-class in README and implement.md

Usage

```bash
npx @luizsantiago/spec-seatbelt loop-plan [feature]
npx @luizsantiago/spec-seatbelt loop-plan auth --json
```

In chat:

```
/loop

Run loop-plan, then implement the next wave. Use sub-agents for parallel groups.
```

v2.0.0 — @luizsantiago/spec-seatbelt

Package rename (breaking)

Before	After
@luizsantiago/agentic-harness	@luizsantiago/spec-seatbelt
agentic-harness CLI	spec-seatbelt CLI

Same GitHub repo (spec-driven-harness), same .specs/ layout. Run install once after switching.

Upgrade

```bash
npx @luizsantiago/spec-seatbelt install
npx @luizsantiago/spec-seatbelt doctor
```

Features (from v1.3.0 lineage)

• doctor, agent-first onboarding, README overhaul
• All phase commands documented (/specify, /loop, /verify, …)
• Task-graph gate, token cost tests, docs/guide/

v1.3.0 — doctor, onboarding, README overhaul

Highlights

• doctor — Harness Ready audit (0–100 score) with prioritized next actions
• Agent-first onboarding — quieter install, .specs/GETTING_STARTED.md, simplified gates map
• README overhaul — install at top, all phase commands documented (/specify, /loop, /verify, /archive, etc.)
• Task-graph gate — validate-tasks requires task-graph.md when 3+ tasks exist on disk
• Token cost tests — ~31k naive dump vs ~9k specify vs ~4k execute (CI-enforced)
• Docs — docs/wiki/ → docs/guide/ (loop patterns, ecosystem, brownfield context, credits)

Upgrade

```bash
npx @luizsantiago/agentic-harness install
npx @luizsantiago/agentic-harness doctor
```

v1.2.1 — revert ChatPRD spike

Reverts v1.2.0 ChatPRD spike (out of scope). Package matches v1.1.x feature set.

Upgrade from 1.2.0: no chatprd CLI or chatprd-sync reference.

v1.2.0 — ChatPRD sync spike

Highlights

• ChatPRD sync (spike) — skills/references/chatprd-sync.md documents MCP auth, one-way PRD→spec workflow, and cloud-agent limits
• chatprd pull --prd-id --dry-run — preview EARS mapping to stdout (requires CHATPRD_API_KEY; use HARNESS_CHATPRD_FIXTURE for local/CI demos)
• Preset placeholders — optional chatprd_org_id in built-in presets; API key via CHATPRD_API_KEY env only
• 18th phase reference in installer catalog

Upgrade

npx @luizsantiago/agentic-harness@1.2.0 install

Out of scope

• Live ChatPRD MCP/HTTP client in CLI (fixture hook only)
• Bidirectional archive → ChatPRD sync

PR: #62

v1.1.0

v1.1.0 — Brownfield project-init

New command: project-init

Map an existing codebase into .specs/ before the first feature:

npx @luizsantiago/agentic-harness project-init --dry-run
npx @luizsantiago/agentic-harness project-init --preset node-ts
npx @luizsantiago/agentic-harness project-init --domains auth,billing

Creates (when missing):

• .specs/project/PROJECT.md — stack + domain map
• .specs/domains/[domain]/spec.md — brownfield stubs
• .specs/project/ROADMAP.md — planned domain drafting
• .specs/config.yaml — auto-detected preset

New reference: references/project-init.md

Install: npx @luizsantiago/agentic-harness@1.1.0 install

v1.0.1

v1.0.1 — Audit fixes

Bug fix

• phase-context no longer leaks unindented YAML comments from init-config stubs into agent output
• Config parser respects YAML literal block indentation rules

Docs / install

• engineering-baseline.mdc gates map synced (archive-feature, phase-context, init-config)
• Install copies templates/config.yaml.example verbatim

npx @luizsantiago/agentic-harness@1.0.1 install

v1.0.0

v1.0.0 — Presets & overrides

Config presets

• Built-in stacks: default, node-ts, python
• preset list / preset show <name>

extends + overrides

.specs/config.yaml can extend a preset and append per-phase rules:

extends: node-ts
context: |
  Team: Acme
overrides:
  rules:
    verify:
      - Also run playwright e2e

CLI

• init-config [--preset <name>]
• install --preset <name>
• phase-context and feature-init resolve the full config chain
• feature-init reads branch_prefix from resolved config

Install

npx @luizsantiago/agentic-harness@1.0.0 install --preset node-ts

v0.9.0

v0.9.0

Archive CLI

• archive-feature — Tier 0 fold-back after Verify PASS
• Updates .specs/project/ROADMAP.md
• Merges delta (ADDED/MODIFIED/REMOVED) or full specs into .specs/domains/[domain]/spec.md
• Resets .specs/STATE.md for the next feature
• Runs validate-state before archive (skip with --skip-verify for recovery only)

Config injection

• phase-context <phase> — prints project context + per-phase rules from .specs/config.yaml

Dev ergonomics

• npm run harness -- <cmd> when developing inside this repo

Install

npx @luizsantiago/agentic-harness@0.9.0 install