v0.10.2 — fix registerTool field name

Renames inputSchema to parameters in all 16 api.registerTool({...}) calls so the OpenClaw plugin loader stops reporting:

plugin tool is malformed (rulespec): <tool> missing parameters object

OpenClaw's host validates factory-returned tools via isRecord(tool.parameters) (openclaw/src/plugins/tools.ts). There is no inputSchema fallback — the field name is load-bearing. The schema shape is unchanged.

v0.10.1 — Dual-skill pattern

v0.10.0 shipped 16 native tools but agents kept reaching for the CLI because the published SKILL.md still documented only npx rulespec. This release fixes that by splitting the skill into two variants.

Changes

• skills/rulespec-plugin/SKILL.md (NEW) — documents the 16 native tools (rulespec_init, rulespec_add_rule, rulespec_emit, …) as the primary interface. Workflow example uses tool calls. rulespec_emit approval-gate config documented.
• skills/rulespec/SKILL.md (existing, lightly amended) — stays the CLI-focused doc for standalone consumers (Claude Code, Cursor, Codex, etc.). Top-of-file note points OpenClaw users at the plugin variant.
• openclaw.plugin.json — skills field now points at ./skills/rulespec-plugin. On openclaw plugins install rulespec, that's the SKILL.md published to ~/.openclaw/plugin-skills/rulespec/SKILL.md.

Both variants ship in the npm tarball — files: ["dist", "skills/**/*"]. The plugin manifest just selects which one OpenClaw publishes on install.

No code changes; same 16 tools register, same plugin behavior, same compat range.

PR: #4

v0.10.0 — Native OpenClaw plugin tools

rulespec is now a real OpenClaw plugin with 16 native tools — agents manage rulespec files without shelling out to the CLI.

Tools

• Setup: rulespec_init, rulespec_set_domain
• Rules: rulespec_list, rulespec_add_rule, rulespec_edit_rule, rulespec_remove_rule
• Sources: rulespec_add_source, rulespec_remove_source
• Examples: rulespec_add_example, rulespec_remove_example, rulespec_add_rule_example, rulespec_remove_rule_example
• Build: rulespec_compile, rulespec_validate, rulespec_emit
• Bulk: rulespec_replace

Approval gate

Only rulespec_emit is gated (it publishes compiled SKILL.md that agents will follow). All other writes run silently. Configurable per host:

"plugins": {
  "entries": {
    "rulespec": {
      "config": {
        "approval": {
          "enabled": true,
          "skipSessionPatterns": ["email"]
        }
      }
    }
  }
}

Compatibility

• Standalone CLI (npx rulespec) and library exports unchanged.
• Plugin install: openclaw plugins install rulespec.

PR: #3

v0.9.0 — OpenClaw plugin layer

Adds an opt-in OpenClaw plugin layer to rulespec. The package keeps working as a standalone CLI + library; this release makes it installable via openclaw plugins install rulespec so the bundled skill ships through the same path as other OpenClaw plugins.

What's new

• openclaw.plugin.json — declares id: "rulespec", publishes skills/rulespec/ (the canonical SKILL.md) into agents' <available_skills>. No contracts.tools; agents continue to use the documented rulespec CLI via exec.
• Plugin entry under src/plugin/index.ts — minimal { id, register } default export. src/index.ts re-exports it so dist/index.js exposes both the plugin entry and the public library surface (loadRules, compileRule, etc.) — standalone consumers see no behavioral change.
• package.json.openclaw — adds extensions, compat: { pluginApi: ">=2026.5.2", minGatewayVersion: ">=2026.5.2" }, build.openclawVersion: "2026.5.12". The files array now includes skills/**/* and openclaw.plugin.json so the npm tarball ships the plugin assets.

Usage

# Standalone (unchanged):
npx rulespec init --domain "..."

# As an OpenClaw plugin:
openclaw plugins install rulespec

PR #1 — #1

v0.8.0 — agent-targeted skill folders

Set the agent location once at init — never repeat it

`rulespec init -a ` drops the skill folder right where that agent looks. Every later command (`add`, `edit`, `emit`, …) auto-discovers it. No flag-juggling on every invocation.

```bash

Project-scoped — committed with your repo

rulespec init --domain payments -a claude-code

→ .claude/skills/payments/

User-global — shared across all projects

rulespec init --domain payments -a claude-code -g

→ ~/.claude/skills/payments/

Default (no -a) unchanged

rulespec init --domain payments

→ skills/payments/

```

Supported agents

`claude-code`, `cursor`, `openclaw`, `codex`, `opencode`, `windsurf`, `amp`, `augment`, `gemini-cli`, `github-copilot`. Path mappings vendored from vercel-labs/skills. Anywhere else: pass `--outdir `.

Per-emit override

You can still retarget a single emit without reconfiguring:

```bash
rulespec emit -a cursor # writes SKILL.md to .agents/skills// instead
```

Other changes

• Discovery now walks every known agent's project + global skills dir
• `emit` writes `SKILL.md` next to the discovered source by default — folder stays self-contained no matter where it lives
• Short-flag support for `-a` (`--agent`) and `-g` (`--global`)
• Clear errors for unknown agent and `--outdir` + `--agent` conflict

Upgrade notes

Fully back-compat. Existing `skills//` folders and legacy `*.rulespec.yaml` at project root still work unchanged.

v0.7.1 — skill-folder scaffolding

Changes

Scaffold each domain as a self-contained skill folder

`rulespec init --domain payments` now creates `skills/payments/rulespec.yaml` (inside the skill folder) instead of a loose `payments.rulespec.yaml` at project root. Running `rulespec emit` writes `SKILL.md` right next to it — so each domain folder is a complete, agent-loadable skill out of the box.

```
skills/
payments/
rulespec.yaml ← authored source
SKILL.md ← emitted (agent-loadable)
```

Auto-discovery now walks skills/*/rulespec.yaml

All CLI commands (`add`, `edit`, `list`, `emit`, …) find your source automatically under the new layout. Legacy `*.rulespec.yaml` at the project root still works — existing projects keep functioning unchanged.

Upgrade notes

No action required. Existing projects continue to work. New `init` invocations use the skill-folder layout.

v0.7.0

Breaking changes:

• rulespec init now requires --domain and creates {domain}.rulespec.yaml (e.g. invoice-processing.rulespec.yaml)
• Example field description renamed to note (--note in CLI)

New features:

• rulespec emit scans all *.rulespec.yaml files and emits all domains at once
• Auto-detection: if only one *.rulespec.yaml exists, all commands use it automatically
• Multiple files: use --file to target a specific one
• add-rule-example and add-example accept --note for context
• Legacy rulespec.yaml still works as fallback

v0.6.0

Clean emitted output for agent prompt injection.

• Moved "do not edit" warning to YAML comment in frontmatter — agents no longer see meta-instructions about CLI usage
• Emitted SKILL.md now contains only rules, sources, and intent tags — optimized for prompt injection
• Two audiences: manager agent gets CLI instructions (skills/rulespec/SKILL.md), worker agent gets clean policy (skills/{domain}/SKILL.md)
• Updated README with agent-specific install examples and file-based examples
• Added ClawHub install path

v0.5.1

Tightened README, updated npm keywords and GitHub topics.

v0.5.0

Updated README with complete CLI reference, npm badges, agent skill install instructions, and proper GitHub topics.