@nareshdama/superskill-policy-engine
v1.0.0
Published
Runner-agnostic policy engine: tool sandbox + approvals, handoff schema, prompt blocks, and human-reviewed proposals for multi-agent IDE/terminal workflows.
Maintainers
Readme
Superskill module
Superskill is a repo-local module that turns skills + scenarios into:
- model settings (temperature mapping with model-family exceptions)
- output contracts (low narration, high signal)
- a strict agent-to-agent handoff packet (schema-validated)
- retry/convergence behavior that reduces roundtrips (policy + optional
retry-planCLI) - trace + outcome logging and human-reviewed policy proposals (no automatic policy writes)
This folder is designed to be vendored into other repos or installed from npm. The package @nareshdama/superskill-policy-engine is published on the public npm registry.
| Where | Link | |-------|------| | GitHub | github.com/nareshdama/Superskill | | npm package | npmjs.com/package/@nareshdama/superskill-policy-engine |
Install from npm
npm i @nareshdama/superskill-policy-engine(npm install @nareshdama/superskill-policy-engine is equivalent.)
Run the CLI without a global install:
npx @nareshdama/superskill-policy-engine --help
npx @nareshdama/superskill-policy-engine compile-prompt --provider openai --environment terminal --intent implement --scenario normalImport the library (Node 18+):
import { /* see docs/PROGRAMMATIC_API.md */ } from "@nareshdama/superskill-policy-engine";Community
| Resource | Purpose | |----------|---------| | CONTRIBUTING.md | Git workflow, dev setup, PR checklist | | Handbook.md | Developer handbook for fusing external cookbooks / OSS patterns into Superskill artifacts | | CHANGELOG.md | Versioned change log (SemVer; current: 1.0.0) | | CODE_OF_CONDUCT.md | Community standards (Contributor Covenant 2.1) | | SECURITY.md | How to report vulnerabilities privately | | Issues | Bugs and feature discussion |
Entry points
- Config:
superskill.yaml(at repo root when using this package directly) - Policies:
policies/ - Schemas:
schemas/(handoff, trace line, outcome line, proposal file, …) - Docs:
docs/— start withdocs/CURRENT_DEVELOPMENT.mdanddocs/ROADMAP.md - CLI:
cli/src/index.mjs(seecli/README.md)
Current status (Phases 1–6 baseline)
Roadmap phases 1 through 6 are implemented at baseline:
- Phase 1–2: policy resolution and prompts; runner-facing validation (
validate-output,validate-tool-args,retry-plan,runner-dry-run). - Phase 3: handoff + trace (
read-handoff,trace-append,trace-tail). - Phase 4: outcomes and
outcomes-report/ proposal artifacts. - Phase 5: tests,
docs/PROGRAMMATIC_API.md,validate-proposal. - Phase 6 (sandbox + approvals): declarative tool policy (
policies/tool-policy.yaml) + runner capability profiles (policies/runner-capabilities.yaml); pre-call gatecheck-tool-call(exit 0/1/2 for auto/ask/deny); prompt blocks injected intocompile-prompt;approvalhandoff block +request-approval/grant-approvalworkflow; canonical gate trace events feeding per-tool aggregates and tightening/loosening proposals.
The CLI does not call model APIs — runners own transport and keys. The gate is read-only; sandboxing is the runner's responsibility, and Superskill refuses to recommend plans that exceed the runner's declared capabilities.
Next steps and out-of-scope items are listed under Future development scope in docs/ROADMAP.md.
Docs (current set)
| Doc | Purpose |
|-----|---------|
| docs/CURRENT_DEVELOPMENT.md | What is implemented now (by phase). |
| docs/ROADMAP.md | Snapshot, roadmap, future dev scope, completed milestones. |
| docs/PHASE4_WORKFLOW.md | Log → report → proposal → PR. |
| docs/HANDOFF_PROTOCOL.md | Handoff packet protocol. |
| docs/TEMPERATURE_LAYER.md | Temperature layer. |
| docs/POLICY_CHANGE_PLAYBOOK.md | Policy changes + resources/skills/. |
| docs/PROGRAMMATIC_API.md | Embeds: lib/ exports. |
| docs/PUBLISHING_NPM.md | npm publishing. |
| docs/TOOL_GATE.md | Sandbox + approvals: check-tool-call, decideToolCall. |
Goals
- Fewer API calls by default (converge via policy, not rerolls)
- Less hallucination (verified pointers + schema gating)
- Maximum actionable output (structured work items)
npm (reference)
- Status: Published — install with
npm i @nareshdama/superskill-policy-engine - Registry: npmjs.com/package/@nareshdama/superskill-policy-engine
- Package name:
@nareshdama/superskill-policy-engine - Binary:
superskill(vianpxor afternpm i/npm install) - Develop in this repo:
npm run smoke,npm test(from clone root)
Publishing workflow: docs/PUBLISHING_NPM.md.
