From 38a588406e2318c55073c8c7ea4e09decbb05fb5 Mon Sep 17 00:00:00 2001 From: Koishore Roy Date: Mon, 8 Jun 2026 17:03:12 +0530 Subject: [PATCH] chore: align positioning to intent-bound action authorization --- .claude-plugin/marketplace.json | 4 ++-- .github/PULL_REQUEST_TEMPLATE.md | 2 +- CONTRIBUTING.md | 2 +- README.md | 10 +++++----- delego/.claude-plugin/plugin.json | 2 +- delego/agents/review/broker-reviewer.md | 6 +++--- delego/agents/review/policy-reviewer.md | 4 ++-- delego/skills/init/SKILL.md | 4 ++-- 8 files changed, 17 insertions(+), 17 deletions(-) diff --git a/.claude-plugin/marketplace.json b/.claude-plugin/marketplace.json index 0428159..40c9cda 100644 --- a/.claude-plugin/marketplace.json +++ b/.claude-plugin/marketplace.json @@ -1,12 +1,12 @@ { "name": "delego", "owner": { "name": "Delego-Dev", "email": "koishore@gmail.com" }, - "description": "delego — a policy & audit firewall for agent actions. Skills, review agents, and the MCP server in one plugin.", + "description": "delego — intent-bound action authorization for AI agents. Skills, review agents, and the MCP server in one plugin.", "plugins": [ { "name": "delego", "source": "./delego", - "description": "Set up and operate the delego firewall: init, policy drafting + adversarial review, approval triage, audit explaining/investigation, and the delego MCP server.", + "description": "Set up and operate delego: init, policy drafting + adversarial review, approval triage, audit explaining/investigation, and the delego MCP server.", "version": "0.1.0" } ] diff --git a/.github/PULL_REQUEST_TEMPLATE.md b/.github/PULL_REQUEST_TEMPLATE.md index 86d255d..5e4cba0 100644 --- a/.github/PULL_REQUEST_TEMPLATE.md +++ b/.github/PULL_REQUEST_TEMPLATE.md @@ -17,7 +17,7 @@ - [ ] Frontmatter is valid and the `description` is **precise** — it fires on the right requests only, and won't hijack unrelated ones. - [ ] Instructions are **safe** (nothing destructive, credential-leaking, or that - bypasses the firewall) and **delego-accurate** (fail-closed, the invariants, + bypasses delego) and **delego-accurate** (fail-closed, the invariants, honest caveats — no overselling the audit). - [ ] I did **not** change the plugin/marketplace manifest or the MCP config (or a maintainer has signed off — these affect every install). diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md index 27ef4e8..fa77a06 100644 --- a/CONTRIBUTING.md +++ b/CONTRIBUTING.md @@ -27,7 +27,7 @@ review it as you would code that executes on someone else's machine. truncation of the most recent receipts, and the policy can't see the URL query string (until protocol 0.3). - **Safe instructions.** No destructive or credential-exfiltrating guidance, and - nothing that bypasses the firewall. + nothing that bypasses delego. - **A precise `description`.** It drives auto-delegation — specific enough to fire on the right requests, not so broad it hijacks unrelated ones. diff --git a/README.md b/README.md index 88208de..e584667 100644 --- a/README.md +++ b/README.md @@ -1,7 +1,7 @@ # delego — Claude Code plugin -The one-command way to get [delego](https://github.com/Delego-Dev/delego) — the -policy & audit firewall for agent actions — into a Claude Code project: its +The one-command way to get [delego](https://github.com/Delego-Dev/delego) — +intent-bound action authorization for AI agents — into a Claude Code project: its **skills**, **review agents**, and the **MCP server**, all in a single plugin. This repo is both the plugin and its own marketplace. @@ -22,7 +22,7 @@ spots (no manual cloning into `.claude/`). Then: ``` `/delego:init` installs the `delego` Python package (`pip install "delego[mcp]"`), -creates the project firewall home at `.claude/.delego`, drafts/validates a policy, +creates the project's delego home at `.claude/.delego`, drafts/validates a policy, and verifies. The plugin's MCP server reads `DELEGO_HOME = /.claude/.delego` and activates once the package is installed and the home exists. @@ -36,7 +36,7 @@ and activates once the package is installed and the home exists. **Skills** (`/delego:`) | Skill | Use it to… | |---|---| -| `init` | Install delego, create the firewall home, verify. | +| `init` | Install delego, create the delego home, verify. | | `policy-drafter` | Draft/harden a fail-closed `policy.yaml`. | | `approval-triage` | Review pending approvals and approve/deny. | | `audit-explainer` | Verify the signed chain and explain what the agent did. | @@ -49,7 +49,7 @@ and activates once the package is installed and the home exists. | `audit-investigator` | Verify the ledger, reconstruct authority paths, flag anomalies. | **MCP server** — `delego_propose_action`, `delego_resolve_action`, -`delego_audit_tail`, `delego_show_policy` (the agent's interface to the firewall). +`delego_audit_tail`, `delego_show_policy` (the agent's interface to delego). ## Typical flow diff --git a/delego/.claude-plugin/plugin.json b/delego/.claude-plugin/plugin.json index f43fddb..855979e 100644 --- a/delego/.claude-plugin/plugin.json +++ b/delego/.claude-plugin/plugin.json @@ -1,7 +1,7 @@ { "name": "delego", "version": "0.1.0", - "description": "Policy & audit firewall for agent actions — skills, review agents, and the MCP server.", + "description": "Intent-bound action authorization for AI agents — skills, review agents, and the MCP server.", "author": { "name": "Koishore Roy", "email": "koishore@gmail.com" }, "homepage": "https://github.com/Delego-Dev/delego", "repository": "https://github.com/Delego-Dev/plugin", diff --git a/delego/agents/review/broker-reviewer.md b/delego/agents/review/broker-reviewer.md index a2ad5e9..53ec323 100644 --- a/delego/agents/review/broker-reviewer.md +++ b/delego/agents/review/broker-reviewer.md @@ -4,7 +4,7 @@ description: Review a custom delego BrokerAdapter for adherence to the core inva tools: Read, Grep, Glob --- -You review delego **BrokerAdapter** implementations. In delego the firewall +You review delego **BrokerAdapter** implementations. In delego the authorization layer *decides*; the broker *executes* the already-authorised action and is the only component that touches a credential. Your job: ensure a broker upholds the invariants and doesn't reopen the holes delego closes. @@ -12,13 +12,13 @@ invariants and doesn't reopen the holes delego closes. ## The contract A broker implements `execute(action: ProposedAction) -> dict`. The `action` carries `method`, `url`, `params`, and the derived `intent_hash` / `fingerprint`. -The firewall has already authorised it; the broker carries it through the +delego has already authorised it; the broker carries it through the component that holds the secret. ## Check — rate critical / high / medium / low 1. **No credential in delego's process.** The upstream secret must live in an external gateway/vault, not be hard-coded or read into a broker running - in-process with the firewall. Forwarding to a gateway that injects it is the + in-process with delego. Forwarding to a gateway that injects it is the pattern; holding the secret in-process is a finding (critical for real creds). 2. **Execute exactly the authorised action.** The request sent must match the fingerprinted `method` / host / `path` / `params`. Flag any place the broker diff --git a/delego/agents/review/policy-reviewer.md b/delego/agents/review/policy-reviewer.md index a1f45e4..1f8aa59 100644 --- a/delego/agents/review/policy-reviewer.md +++ b/delego/agents/review/policy-reviewer.md @@ -4,8 +4,8 @@ description: Adversarially review a delego policy.yaml — fail-closed correctne tools: Read, Grep, Glob, Bash --- -You are a delego **policy reviewer**. delego is a deterministic policy & audit -firewall for agent actions: a policy decides whether an agent's proposed action +You are a delego **policy reviewer**. delego is deterministic, intent-bound +action authorization for AI agents: a policy decides whether an agent's proposed action is allowed, parked for human approval, or denied — fail-closed, with no LLM in the decision. Your job is to find every way a policy is too loose, inconsistent with its stated intent, or footgunned, and report concrete fixes. diff --git a/delego/skills/init/SKILL.md b/delego/skills/init/SKILL.md index f9d8c4f..81f9bda 100644 --- a/delego/skills/init/SKILL.md +++ b/delego/skills/init/SKILL.md @@ -1,6 +1,6 @@ --- name: init -description: Initialize delego — the policy & audit firewall for agent actions — in the current project. Installs the package, creates a project-scoped firewall home with signing keys and a starter policy, and verifies it. The delego MCP server is provided by this plugin, so this does NOT touch .mcp.json. Use when the user wants to "set up delego", "add delego to this project", or "initialize the delego firewall". +description: Initialize delego — intent-bound action authorization for AI agents — in the current project. Installs the package, creates a project-scoped delego home with signing keys and a starter policy, and verifies it. The delego MCP server is provided by this plugin, so this does NOT touch .mcp.json. Use when the user wants to "set up delego", "add delego to this project", or "initialize delego". --- # init @@ -34,7 +34,7 @@ Require Python ≥ 3.10. `delego-mcp` is on the PATH Claude Code launches the MCP server with (activate the venv, or install into the environment Claude Code uses). -2. **Create the firewall home** the plugin's MCP server points at: +2. **Create the delego home** the plugin's MCP server points at: ```bash .venv/bin/delego --home .claude/.delego init ```