Skip to content
Agentmart
Submit a tool

threadlight-governed-actions

Use when a pilot needs consequential-action runtime implementation or evidence: selected native hooks, gateway effect closure, action inventory, pre-deploy…

Skill by aiappsgbb

IndexedMIT★ 2
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/aiappsgbb-threadlight-governed-actions

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Use when a pilot needs consequential-action runtime implementation or evidence: selected native hooks, gateway effect closure, action inventory, pre-deploy enforcement gates, approval anti-replay, payload-free audit, Governance Evidence Pack or GHCP change-plane controls. Not for quality evals, model content filtering or production certification.

Use cases: devops, security, writing

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 26, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found
Runs as
Python package

Network hosts it talks to 3

  • codeload.github.com · found in install command
  • github.com · found in code
  • threadlight.dev · found in code

Files it touches

  • ~/.claude/skills/threadlight-governed-actions

What's inside

Skill name
threadlight-governed-actions
Description
Use when a pilot needs consequential-action runtime implementation or evidence: selected native hooks, gateway effect closure, action inventory, pre-deploy enforcement gates, approval anti-replay, payload-free audit, Governance Evidence Pack or GHCP change-plane controls. Not for quality evals, model content filtering or production certification.
License
MIT
Metadata
version: 2.0.0
Instructions
380 lines in SKILL.md · view on GitHub

Before installing

  • Ships executable scripts

Files (122)

  • SKILL.md
  • finding-catalog.json · references/
  • governed-actions-apply-plan.schema.json · references/
  • governed-actions-manifest.schema.json · references/
  • agent_hooks_interceptor.py.tmpl · references/scaffold/
  • approval-binding-fixture.json.tmpl · references/scaffold/
  • governed-actions.policy.yaml.tmpl · references/scaffold/
  • governed-actions.yml.tmpl · references/scaffold/
  • test_governed_actions_contract.py.tmpl · references/scaffold/
  • upstream-pin.json · references/
  • upstream-pin.md · references/
  • __init__.py · scripts/
  • alerts.py · scripts/
  • canonical.py · scripts/
  • contracts.py · scripts/
  • ghcp.py · scripts/
  • governed_actions.py · scripts/
  • inputs.py · scripts/
  • inventory.py · scripts/
  • maf_adapter.py · scripts/
  • mediation.py · scripts/
  • native_local.py · scripts/
  • native_local_child.py · scripts/
  • probes.py · scripts/
  • render.py · scripts/
  • scaffold.py · scripts/
  • __init__.py · tests/
  • conftest.py · tests/
  • CODEOWNERS · tests/fixtures/approval-replay/.github/
  • governed-actions.yml · tests/fixtures/approval-replay/.github/workflows/
  • agent.yaml · tests/fixtures/approval-replay/
  • agent.py · tests/fixtures/approval-replay/app/
  • alerts.json · tests/fixtures/approval-replay/governance/
  • change-plane.json · tests/fixtures/approval-replay/governance/
  • installed-packages.json · tests/fixtures/approval-replay/governance/
  • probe-contract.json · tests/fixtures/approval-replay/governance/
  • SPEC.md · tests/fixtures/approval-replay/specs/
  • CODEOWNERS · tests/fixtures/conformant-maf/.github/
  • governed-actions.yml · tests/fixtures/conformant-maf/.github/workflows/
  • agent.yaml · tests/fixtures/conformant-maf/
  • agent.py · tests/fixtures/conformant-maf/app/
  • alerts.json · tests/fixtures/conformant-maf/governance/
  • change-plane.json · tests/fixtures/conformant-maf/governance/
  • installed-packages.json · tests/fixtures/conformant-maf/governance/
  • probe-contract.json · tests/fixtures/conformant-maf/governance/
  • SPEC.md · tests/fixtures/conformant-maf/specs/
  • CODEOWNERS · tests/fixtures/interceptor-failure/.github/
  • governed-actions.yml · tests/fixtures/interceptor-failure/.github/workflows/
  • agent.yaml · tests/fixtures/interceptor-failure/
  • agent.py · tests/fixtures/interceptor-failure/app/
  • alerts.json · tests/fixtures/interceptor-failure/governance/
  • change-plane.json · tests/fixtures/interceptor-failure/governance/
  • installed-packages.json · tests/fixtures/interceptor-failure/governance/
  • probe-contract.json · tests/fixtures/interceptor-failure/governance/
  • SPEC.md · tests/fixtures/interceptor-failure/specs/
  • CODEOWNERS · tests/fixtures/output-streaming/.github/
  • governed-actions.yml · tests/fixtures/output-streaming/.github/workflows/
  • agent.yaml · tests/fixtures/output-streaming/
  • agent.py · tests/fixtures/output-streaming/app/
  • alerts.json · tests/fixtures/output-streaming/governance/
  • …and 62 more

SKILL.md (first lines)

# Threadlight Governed Actions — implement and prove selected action paths

An agent that can only read is a demo risk. An agent that can *write*, *egress*,
or do something *irreversible* is a production risk, and the question a customer
actually asks is narrow: **when this agent takes a consequential action, what
stops it, who approved it, and where is the record?**

For implementation requests, produce actual selected runtime templates using
[generate.py](../threadlight-deploy/references/governance/generate.py) and its
[configuration/ordering contract](../threadlight-deploy/references/governance/README.md).
Use `threadlight-govern` for the real ACS bundle and authorized signed distribution;
implement the host-owned application seam, not an allow-only test adapter.
Assessment alone is not implementation. Deployment remains separately authorized.

For [signed business evidence](../../docs/signed-evidence.md), the assessor retains
`signed-evidence-proof-required` as `not-verified`: generic probes, LOCAL-14 and
hosted noop receipts cannot establish an issuer/profile/source-revision contract.
Catalog JWT/MCP fixtures do not close a customer's business acceptance gate.

For consent by the **requesting user**, `user-confirmation-proof-required` also
remains `not-verified`: generic approval probes cannot prove matching user
identity, the selected provider's assurance, a real out-of-band decision or the
deployment's one-use confirmation flow. It is distinct from third-party review.
Use the actual confirmation runtime and retain separate target-bound acceptance;
an email, local test or noop receipt is not MFA or business-write proof.

The **assessor CLI** is read-only by default and renders a deterministic manifest,
evidence pack and never-self-applying remediation plan. These are selected-path
evidence, **not whole-agent** governance. SAFE is the method, ACS/Rego the PDP,
Agent Hooks the host/interceptor contract, native host/gateway the PEP, AGT the
toolkit, and ASSERT assurance.

### Native selected local proof

The current `LOCAL-14` producer runs the actual served application/native path
and returns assessor-verified allow/deny/failure/approval/terminal-effect proof.
The standalone export includes `.governance-tools`, exact-pin tooling, official
CTK sources/test-oracle build inputs and native-local CI. Follow the generated
commands; do not depend on a catalog checkout accidentally on `PYTHONPATH`.
Native/CTK uses published unmodified execution packages: 47 declared vectors,
four undeclared incremental-output vectors and only a corrected test oracle.
Native APIs are preview/alpha, with explicit experimental defaults.

Unbound reads remain unbound and never invoke ACS. Selected bindings require
signed fresh policy, host-trusted facts, authenticated one-use human approval
when required, and central audit ACK before effects. Recheck after awaits at
the actual terminal target; a check at function start is insufficient. GHCP
supports registered gateway effects only, not arbitrary URLs/shells or full
lifecycle/output coverage. `enforced`, `observed`, `unbound`, `unverified`,
`unsupported` and `bypassable` are binding statuses, not whole-agent verdicts.

Local proof remains local. The Task11 collector only invokes the reserved,
registered `governance_probe_noop`; it cannot certify business bindings.
`returns_apply_decision` has local Cosmos decision/audit proof, not live payment
or settlement proof. A new deployment requires fresh after-deploy evidence;
policy/CI or legacy v2 green cannot replace it.

```
declared actions ─┐
implemented code ─┼─► inventory ─► mediation graph ─► application probes ─┐
policy/approval  ─┘                                                       ├─► manifest
workflows/CODEOWNERS ─────────────► change-plane assessment ──────────────┤   evidence pack
alert catalog ────────────────────► alert posture ────────────────────────┘   apply plan
```

## When to invoke

| Phase | Invoke | What it proves |
| --- | --- | --- |
| `design` | after `threadlight-design`, before code exists | the SPEC § 8 action declaration is complete and every action has an explicit consequence class |
| `pre-deploy` | before `threadlight-deploy` | declared and implemented actions agree; path-bound allow/deny receipts verify mediation candidates; separate generic probes exercise deny/transform/fail-closed, approval anti-replay, output gating, and payload-free audit |
| `post-deploy` | **staging only** | the same assessment against a deployed non-production environment; `--phase post-deploy` refuses to run without `--staging-resource-group` naming an explicit, non-production staging resource group |

`post-deploy` is deliberately staging-only. This skill never runs, not even
read-only, against an unnamed or production environment.

## Inputs

Everything is repository-relative and read-only. Nine categories, each an
explicit rule — discovery never widens into "anything under the repository":

| Input | Where | Used by |
| --- | --- | --- |
| SPEC action declaration | `specs/SPEC.md` § 8 (required for `design`/`pre-deploy`) | `ACT-001`, `ACT-002` |
| Action registries | `agent.yaml`, `agent.yml`, `tool-registry.json`, `tool_registry.json` (target root) | `ACT-001`, `ACT-002` |
| Runtime + middleware source | every non-vendored `*.py` under the target | `MED-001`..`MED-003`, `ENF-001`, `ENF-002` |
| Policy files | `governance/**/*.{json,yaml,yml}`, `policies/**/*.{json,yaml,yml}` | `MED-001`, `ENF-001` |
| Approval handling | runtime files whose AST references `issue_approval`, `consume_approval`, `redeem_approval`, or `ApprovalHandler` | `APR-001` |
| Tests, conformance, evals | `tests/**`, `**/conformance/**`, `**/evals/**` | `ENF-001`, `GHCP-003`, `PIN-001` |
| Workflows | `.github/workflows/*.yml`, `*.yaml` (direct children) | `GHCP-001`, `GHCP-003`..`GHCP-006` |
| Ownership | `CODEOWNERS` and `.github/CODEOWNERS` | `GHCP-002` |
| Dependency pins | `pyproject.toml`, `requirements*.txt`, lockfiles | `PIN-001` |
| Alert catalog | `governance/alerts.json` | `OPS-001` |
| Optional live evidence | `--live-github` (`gh api`), `--subscription`/`--deploy-identity` (`az ... list`) | `GHCP-002`, `GHCP-006` |

Optional live evidence is exactly that: **optional**. When it is not selected, or
when a permission is missing, the affected control is reported `not-verified`.
A `not-verified` finding is never silently converted to `pass`.

## Path-bound execution contract

`governance/probe-contract.json` must declare one importable application
execution dispatcher and bind every assessor-discovered, non-provider mediation
path to its recomputed path id and action/mode routing-table key:

```json
{
  "dispatch": "app.agent:dispatch",
  "execution_dispatch": "app.agent:dispatch_execution_path",
  "execution_paths": [
    {
      "action_id": "payments.refund",
      "mode": "background",
      "path_id": "400019bbedc16b75"
    }
  ]
}
```

From aiappsgbb/threadlight-skills, MIT.

More from aiappsgbb/threadlight-skills

Signals

Stars
2
Forks
5
Open issues
5
Last commit (this folder)
Oct 5, 2026 (24 days ago)

From GitHub (fetched Oct 24, 2026).

Details

Version
2.0.0
License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from aiappsgbb

Cite this page

threadlight-governed-actions skill. Agent Mart. https://agentmart.dev/skills/aiappsgbb-threadlight-governed-actions (last updated 2026-10-24)