@probo/skills 0.1.0

package/.agents/plugins/marketplace.json

@@ -0,0 +1,30 @@
+{
+  "name": "probo",
+  "interface": {
+    "displayName": "Probo"
+  },
+  "plugins": [
+    {
+      "name": "probo",
+      "description": "Open-source compliance workflows powered by Probo MCP",
+      "source": {
+        "source": "local",
+        "path": "./"
+      },
+      "policy": {
+        "installation": "AVAILABLE",
+        "authentication": "ON_INSTALL"
+      },
+      "category": "Productivity",
+      "homepage": "https://github.com/getprobo/probo/tree/main/packages/skills",
+      "repository": "https://github.com/getprobo/probo",
+      "license": "MIT",
+      "keywords": [
+        "probo",
+        "compliance",
+        "grc",
+        "mcp"
+      ]
+    }
+  ]
+}

package/.claude-plugin/marketplace.json

@@ -0,0 +1,28 @@
+{
+  "name": "probo",
+  "description": "Probo Claude Code plugins",
+  "owner": {
+    "name": "Probo Inc",
+    "email": "hello@probo.com"
+  },
+  "plugins": [
+    {
+      "name": "probo",
+      "description": "Open-source compliance workflows powered by Probo MCP",
+      "source": {
+        "source": "npm",
+        "package": "@probo/skills"
+      },
+      "homepage": "https://github.com/getprobo/probo/tree/main/packages/skills",
+      "repository": "https://github.com/getprobo/probo",
+      "license": "MIT",
+      "keywords": [
+        "probo",
+        "compliance",
+        "grc",
+        "mcp",
+        "agent-skills"
+      ]
+    }
+  ]
+}

package/.claude-plugin/plugin.json

@@ -0,0 +1,21 @@
+{
+  "name": "probo",
+  "version": "0.1.0",
+  "description": "Open-source compliance workflows powered by Probo MCP (Claude Code, Codex, OpenCode)",
+  "author": {
+    "name": "Probo Inc",
+    "email": "hello@probo.com"
+  },
+  "homepage": "https://github.com/getprobo/probo/tree/main/packages/skills",
+  "repository": "https://github.com/getprobo/probo",
+  "license": "MIT",
+  "keywords": [
+    "probo",
+    "compliance",
+    "grc",
+    "mcp",
+    "open-source"
+  ],
+  "skills": "./skills/",
+  "mcpServers": "./.mcp.json"
+}

package/.codex-plugin/plugin.json

@@ -0,0 +1,21 @@
+{
+  "name": "probo",
+  "version": "0.1.0",
+  "description": "Open-source compliance workflows powered by Probo MCP (Codex, OpenCode, Claude Code)",
+  "author": {
+    "name": "Probo Inc",
+    "email": "hello@probo.com"
+  },
+  "homepage": "https://github.com/getprobo/probo/tree/main/packages/skills",
+  "repository": "https://github.com/getprobo/probo",
+  "license": "MIT",
+  "keywords": [
+    "probo",
+    "compliance",
+    "grc",
+    "mcp",
+    "open-source"
+  ],
+  "skills": "./skills/",
+  "mcpServers": "./.mcp.json"
+}

package/.mcp.json

@@ -0,0 +1,8 @@
+{
+  "mcpServers": {
+    "probo": {
+      "type": "http",
+      "url": "${PROBO_BASE_URL}/mcp/v1"
+    }
+  }
+}

package/COMPATIBILITY.md

@@ -0,0 +1,170 @@
+<!--
+Copyright (c) 2026 Probo Inc <hello@probo.com>.
+
+Permission to use, copy, modify, and/or distribute this software for any
+purpose with or without fee is hereby granted, provided that the above
+copyright notice and this permission notice appear in all copies.
+
+THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH
+REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT,
+INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR
+OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+PERFORMANCE OF THIS SOFTWARE.
+-->
+
+# Multi-agent compatibility
+
+`@probo/skills` targets **Claude Code**, **Codex**, **OpenCode**, and
+other MCP-capable agents (including **Cursor**). The portable core is Probo
+MCP plus Agent Skills–compatible `SKILL.md` files.
+
+## What works where
+
+| Component | Claude Code | Codex | OpenCode | Cursor |
+| --- | --- | --- | --- | --- |
+| Probo MCP (OAuth) | ✅ Plugin `.mcp.json` | ✅ `.codex-plugin` + `.mcp.json` | ✅ Manual MCP config | ✅ IDE MCP settings |
+| Skills (`SKILL.md`) | ✅ `skills/` | ✅ `skills/` via `.codex-plugin` | ✅ `.opencode/skills/` or `.claude/skills/` | ✅ Copy/symlink to `.cursor/skills/` |
+| Commands | ✅ `commands/` → `/probo:…` | ⚠️ Use skills instead | ⚠️ Native `skill` tool | ❌ Use skill or rules |
+| Plugin manifest | `.claude-plugin/` | `.codex-plugin/` | Discovery paths (no manifest) | No native manifest |
+| Marketplace catalog | `.claude-plugin/marketplace.json` (repo root or package) | `.agents/plugins/marketplace.json` (repo root or package) | — | — |
+
+## Probo MCP (all agents)
+
+Set the instance URL:
+
+```bash
+export PROBO_BASE_URL="https://your-probo-instance.example.com"
+```
+
+Endpoint: `${PROBO_BASE_URL}/mcp/v1` (HTTP, OAuth 2.0). Do not configure a
+static bearer token — OAuth discovery uses
+`/.well-known/oauth-protected-resource`.
+
+### Claude Code
+
+**From the monorepo or GitHub** (repo-root catalog at
+`.claude-plugin/marketplace.json`):
+
+```bash
+claude plugin marketplace add getprobo/probo
+# or, from a local clone:
+claude plugin marketplace add .
+claude plugin install probo@probo
+claude mcp login probo   # or /mcp in session
+/probo:access-review Q3 GitHub review
+```
+
+**From the package directory** (catalog resolves `@probo/skills` from npm):
+
+```bash
+claude plugin marketplace add ./packages/skills/.claude-plugin
+claude plugin install probo@probo
+claude mcp login probo
+```
+
+Or install the plugin directory directly:
+
+```bash
+claude --plugin-dir ./packages/skills
+```
+
+### Codex
+
+**From the monorepo or GitHub** (repo-root catalog at
+`.agents/plugins/marketplace.json`):
+
+```bash
+codex plugin marketplace add getprobo/probo
+# or, from a local clone:
+codex plugin marketplace add .
+codex plugin install probo@probo
+codex mcp login probo
+```
+
+**From the package directory** (catalog at
+`packages/skills/.agents/plugins/marketplace.json`):
+
+```bash
+codex plugin marketplace add ./packages/skills
+codex plugin install probo@probo
+codex mcp login probo
+```
+
+Or install the plugin directory directly:
+
+```bash
+codex plugin install ./packages/skills
+codex mcp login probo
+```
+
+Skills load from `./skills/` via `.codex-plugin/plugin.json`. The repo-root
+marketplace `source.path` is `./packages/skills`; the package-level
+catalog uses `./` (plugin package root).
+
+### OpenCode
+
+OpenCode discovers skills at `.opencode/skills/`, `.claude/skills/`, and
+`~/.config/opencode/skills/`. Options:
+
+**Option A — symlink from this package:**
+
+```bash
+mkdir -p .opencode/skills
+ln -s ../../packages/skills/skills/access-review .opencode/skills/access-review
+ln -s ../../packages/skills/skills/open-source-compliance .opencode/skills/open-source-compliance
+```
+
+**Option B — Claude Code bridge:** install
+[`opencode-claude-code-bridge`](https://www.npmjs.com/package/opencode-claude-code-bridge)
+to import Claude plugins and MCP configs into OpenCode.
+
+Configure Probo MCP in `opencode.json` or global OpenCode MCP settings, then
+authenticate. Invoke via the native `skill` tool (`access-review`).
+
+### Cursor
+
+1. Add Probo MCP in Cursor settings (HTTP URL: `${PROBO_BASE_URL}/mcp/v1`,
+   OAuth).
+2. Copy or symlink skills into `.cursor/skills/`:
+
+```bash
+mkdir -p .cursor/skills
+cp -r packages/skills/skills/access-review .cursor/skills/
+```
+
+Reference the skill in chat or add a Cursor rule pointing at the skill.
+
+## Portable vs agent-specific paths
+
+| Path | Portable? |
+| --- | --- |
+| `skills/<name>/SKILL.md` | ✅ Agent Skills standard |
+| `skills/<name>/references/*.md` | ✅ Relative to skill directory |
+| `.mcp.json` with `${PROBO_BASE_URL}` | ✅ Standard env var |
+| `${CLAUDE_PLUGIN_ROOT}` | ❌ Claude Code only — avoid in skill bodies |
+| `commands/*.md` | Claude Code slash commands only |
+
+Skill bodies use **relative** `references/` paths so they work once the skill
+directory is discovered, regardless of which agent loads it.
+
+## npm package layout
+
+```
+@probo/skills/
+  .claude-plugin/plugin.json         # Claude Code manifest
+  .claude-plugin/marketplace.json    # Claude marketplace (npm)
+  .codex-plugin/plugin.json          # Codex manifest
+  .agents/plugins/marketplace.json   # Codex marketplace (package-local)
+  .mcp.json                          # Shared MCP wiring
+  skills/                            # Shared skills (all agents)
+  commands/                          # Claude Code commands only
+```
+
+Repo root (monorepo / `getprobo/probo` Git installs):
+
+```
+.claude-plugin/marketplace.json      # Claude marketplace → packages/skills
+.agents/plugins/marketplace.json     # Codex marketplace → packages/skills
+```

package/README.md

@@ -0,0 +1,96 @@
+# @probo/skills
+
+Multi-agent compliance skills for open-source GRC workflows. Ships Agent
+Skills–compatible instructions and wires agents to the [Probo MCP
+API](https://github.com/getprobo/probo/tree/main/pkg/server/api/mcp/v1) via
+OAuth 2.0.
+
+**Supported agents:** Claude Code, Codex, OpenCode, Cursor (MCP + skills).
+
+Marketplace catalogs: `.claude-plugin/marketplace.json` at the repo root or
+under `packages/skills/` (Claude Code), `.agents/plugins/marketplace.json` at
+the repo root or under `packages/skills/` (Codex). See
+[COMPATIBILITY.md](./COMPATIBILITY.md).
+
+## Install
+
+### From GitHub or npm
+
+**From GitHub** (repo-root catalog at `.claude-plugin/marketplace.json`):
+
+```bash
+claude plugin marketplace add getprobo/probo
+# or, from a local clone:
+claude plugin marketplace add .
+claude plugin install probo@probo
+```
+
+**From the package directory** (catalog at
+`packages/skills/.claude-plugin/marketplace.json`, resolves `@probo/skills`
+from npm):
+
+```bash
+claude plugin marketplace add ./packages/skills/.claude-plugin
+claude plugin install probo@probo
+```
+
+When consuming the published package, the package-level marketplace entry
+resolves `@probo/skills` from npm (see
+`packages/skills/.claude-plugin/marketplace.json`).
+
+### Configure Probo MCP
+
+Set your Probo instance URL before starting Claude Code:
+
+```bash
+export PROBO_BASE_URL="https://your-probo-instance.example.com"
+```
+
+The plugin `.mcp.json` connects to `${PROBO_BASE_URL}/mcp/v1`. Probo MCP
+authenticates with **OAuth 2.0** — no API token or bearer header is required in
+the plugin config. On first use, sign in from Claude Code:
+
+```text
+/mcp
+```
+
+Or from your shell:
+
+```bash
+claude mcp login probo
+```
+
+Claude Code discovers Probo's authorization server via
+`/.well-known/oauth-protected-resource` and stores tokens securely.
+
+### Local development
+
+```bash
+claude --plugin-dir ./packages/skills
+```
+
+## What's included
+
+| Component | Location | Purpose |
+| --- | --- | --- |
+| MCP | `.mcp.json` | Probo API connection |
+| Skills | `skills/` | Compliance workflows |
+| Commands | `commands/` | `access-review`, `missing-signatures` — semi-auto workflows |
+
+Skills: `/probo:<skill-name>` (e.g. `/probo:open-source-compliance`, `/probo:missing-signatures`).
+
+Commands: `/probo:<command-name>` (e.g. `/probo:access-review`, `/probo:missing-signatures`).
+
+## Adding content
+
+See [`contrib/claude/skills.md`](../../contrib/claude/skills.md).
+
+```bash
+npm --workspace @probo/skills run validate
+claude --plugin-dir ./packages/skills
+```
+
+## Release
+
+Published to npm as `@probo/skills`. See
+[`contrib/claude/release/skills.md`](../../contrib/claude/release/skills.md).

package/commands/access-review.md

@@ -0,0 +1,17 @@
+---
+description: Run a semi-automated access review on a Probo campaign (Claude Code slash command).
+argument-hint: [campaign name or id]
+disable-model-invocation: true
+---
+
+# Access review command
+
+Execute the `access-review` skill for campaign `$ARGUMENTS`.
+
+1. Load `skills/access-review/SKILL.md` from the plugin package root.
+2. Load reference docs from `skills/access-review/references/` as directed by
+   the skill.
+3. Follow the skill workflow exactly.
+
+Do not duplicate skill logic here — the skill is the canonical workflow shared
+with Codex and OpenCode.

package/commands/missing-signatures.md

@@ -0,0 +1,17 @@
+---
+description: Report missing document signatures and pending quorum approvals per person (Claude Code slash command).
+argument-hint: [organization name or id]
+disable-model-invocation: true
+---
+
+# Missing signatures command
+
+Execute the `missing-signatures` skill for organization `$ARGUMENTS`.
+
+1. Load `skills/missing-signatures/SKILL.md` from the plugin package root.
+2. Load reference docs from `skills/missing-signatures/references/` as directed
+   by the skill.
+3. Follow the skill workflow exactly.
+
+Do not duplicate skill logic here — the skill is the canonical workflow shared
+with Codex and OpenCode.

package/package.json

@@ -0,0 +1,49 @@
+{
+  "name": "@probo/skills",
+  "version": "0.1.0",
+  "description": "Multi-agent compliance skills for open-source GRC (Claude Code, Codex, OpenCode) powered by Probo MCP",
+  "private": false,
+  "scripts": {
+    "validate": "node scripts/validate.mjs",
+    "release": "npm run validate && npm publish --access public",
+    "prepublishOnly": "npm run validate"
+  },
+  "author": {
+    "name": "Probo Inc",
+    "email": "hello@probo.com"
+  },
+  "files": [
+    ".claude-plugin",
+    ".codex-plugin",
+    ".agents",
+    "skills",
+    "commands",
+    ".mcp.json",
+    "README.md",
+    "COMPATIBILITY.md"
+  ],
+  "keywords": [
+    "claude-code",
+    "agent-skills",
+    "codex",
+    "opencode",
+    "probo",
+    "compliance",
+    "grc",
+    "mcp",
+    "open-source"
+  ],
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/getprobo/probo.git",
+    "directory": "packages/skills"
+  },
+  "bugs": {
+    "url": "https://github.com/getprobo/probo/issues"
+  },
+  "homepage": "https://github.com/getprobo/probo/tree/main/packages/skills#readme",
+  "publishConfig": {
+    "access": "public"
+  }
+}

package/skills/access-review/SKILL.md

@@ -0,0 +1,85 @@
+---
+name: access-review
+description: Run a semi-automated Probo access review campaign. Use when the user wants to review access entries, decide approve/revoke/escalate, or resume an in-progress campaign with MCP and .probo/access-reviews/ notes.
+compatibility: Requires Probo MCP (OAuth 2.0) and file write access for .probo/access-reviews/
+---
+
+# Access review
+
+Run a **semi-automated** access review for campaign `$ARGUMENTS` (or ask the
+user for the campaign name). Review entries only — do not create, start,
+cancel, or close campaigns.
+
+Before executing, read these files **relative to this skill directory**:
+
+- `references/mcp-tools.md` — MCP tool names, inputs, pagination
+- `references/decision-rubric.md` — semi-auto decision rules
+- `references/notes-format.md` — working memory file schema
+
+## Preconditions
+
+1. Probo MCP must be connected. If tools fail with auth errors, stop and tell
+   the user to complete OAuth sign-in for the Probo MCP server in their agent
+   (Claude Code: `/mcp` or `claude mcp login probo`; Codex: `codex mcp login
+   probo`; OpenCode/Cursor: configure MCP in settings then authenticate).
+2. Resolve the campaign from `$ARGUMENTS` (name match or GID). If ambiguous,
+   list `listAccessReviewCampaigns` results and ask the user to pick one.
+3. Campaign `status` must be `IN_PROGRESS` or `PENDING_ACTIONS`. Stop with a
+   clear message for `DRAFT`, `COMPLETED`, or `CANCELLED`.
+
+## Working notes file
+
+Create or resume `.probo/access-reviews/<campaign-slug>.md` per
+`references/notes-format.md`. Create `.probo/access-reviews/` if missing.
+
+## Workflow
+
+### 1. Orient
+
+- Call `getAccessReviewStatistics` for the campaign.
+- Summarize totals and pending count for the user.
+- If no pending entries, report completion and stop.
+
+### 2. Fetch batch
+
+- Call `listAccessEntries` with `campaign_id`, `filter.decision: PENDING`,
+  `size: 50`.
+- Use `last_cursor` from the notes file when resuming.
+
+### 3. Classify each entry
+
+Apply `references/decision-rubric.md`:
+
+| Class | Action |
+| --- | --- |
+| **Auto** | Queue for `recordAccessReviewEntryDecisions` |
+| **Ambiguous** | Present to user; do not write yet |
+| **Skip** | Log in notes only (no API write) |
+
+Hold auto decisions in memory until step 4 succeeds — do not append them to the
+notes file yet.
+
+### 4. Write auto decisions
+
+- Batch via `recordAccessReviewEntryDecisions` when possible.
+- Non-`APPROVED` decisions **must** include `decision_note`.
+- On MCP error, stop and do not advance `last_cursor` or update entry notes.
+- After a successful API response, append each recorded auto decision to the
+  notes file `## Entry notes` table.
+
+### 5. Present ambiguous entries
+
+Show email, roles, flags, proposed decision, rationale. Record only after
+explicit user confirmation.
+
+### 6. Checkpoint
+
+Update notes: `last_cursor`, session log, `updated_at`. Ask to continue if
+`next_cursor` is set.
+
+## Hard rules
+
+- Never call `closeAccessReviewCampaign` or campaign setup mutations unless
+  the user explicitly requests setup work outside this skill.
+- Never invent entry IDs or decisions — use MCP responses only.
+- Never record non-`APPROVED` without `decision_note`.

package/skills/access-review/references/decision-rubric.md

@@ -0,0 +1,60 @@
+# Semi-auto decision rubric
+
+Classify each `PENDING` entry before writing to Probo.
+
+## Auto — record without asking
+
+Apply the **first matching rule** (top to bottom). Always set `decision_note`
+for non-`APPROVED` decisions.
+
+| Condition | Decision | decision_note template |
+| --- | --- | --- |
+| `TERMINATED_USER` flag | `REVOKE` | Terminated user — access no longer required |
+| `CONTRACTOR_EXPIRED` flag | `REVOKE` | Contractor engagement ended |
+| `active === false` and no `NEW` tag | `REVOKE` | Account inactive at source |
+| `ORPHANED` flag and `active === false` | `REVOKE` | Orphaned inactive account |
+| `ORPHANED` flag only, `active !== false` | `ESCALATE` | Orphaned account still active — needs owner |
+| `SHARED_ACCOUNT` flag | `ESCALATE` | Shared account — assign individual owner |
+| `SOD_CONFLICT` flag | `ESCALATE` | Segregation of duties conflict |
+| `PRIVILEGED_ACCESS` or `ROLE_CREEP` flag | `ESCALATE` | Privileged access requires explicit approval |
+| `is_admin === true` and (`DORMANT` flag or last_login very stale) | `ESCALATE` | Admin access dormant — confirm business need |
+| `account_type === SERVICE_ACCOUNT`, active, no danger flags | `APPROVED` | Service account with expected access |
+| Active user, no flags (or only `NONE`), not admin, `incremental_tag !== NEW` | `APPROVED` | Routine access reaffirmed |
+
+"Very stale" last_login: no login in 90+ days when `last_login` is present.
+
+## Ambiguous — show user, do not write
+
+| Condition | Suggested default | Why ambiguous |
+| --- | --- | --- |
+| `incremental_tag === NEW` | `ESCALATE` | New access since last campaign |
+| `NO_BUSINESS_JUSTIFICATION` flag | `ESCALATE` or `REVOKE` | Needs human judgment |
+| `OUT_OF_DEPARTMENT` flag | `ESCALATE` | Role/department mismatch |
+| `EXCESSIVE` or `ROLE_MISMATCH` flag | `ESCALATE` | Role change needs context |
+| `is_admin === true` without dormant signals | `ESCALATE` | Admin approvals need explicit sign-off |
+| `mfa_status === DISABLED` and (`is_admin` or privileged flags) | `ESCALATE` | MFA gap on sensitive access |
+| `auth_method === API_KEY` or `PASSWORD` on production-like roles | `ESCALATE` | Non-SSO auth on sensitive access |
+| Multiple conflicting flags | `ESCALATE` | Rubric rules disagree |
+| `active === null` with revoke-leaning flags | `ESCALATE` | Unknown activity state |
+
+Present the suggested decision; wait for explicit user confirmation.
+
+## Flagging before decision
+
+Do not auto-flag unless the user asks. When reviewing ambiguous entries, you
+may **suggest** `flagAccessReviewEntry` if Probo shows `NONE` but signals are
+obvious (e.g. admin + 180d no login → suggest `DORMANT`).
+
+## Decision notes
+
+- `APPROVED` — `decision_note` optional
+- `REVOKE`, `DEFER`, `ESCALATE` — `decision_note` **required** (MCP rejects empty)
+- Keep notes short, factual, auditable. Reference flags and activity signals.
+
+## DEFER vs ESCALATE
+
+- `ESCALATE` — needs another reviewer or manager (security, HR, app owner)
+- `DEFER` — modify access (role change, downgrade) before final approval; use
+  when the user indicates access should change rather than fully revoke
+
+Default to `ESCALATE` when unsure.

package/skills/access-review/references/mcp-tools.md

@@ -0,0 +1,93 @@
+# Access review MCP tools
+
+All tools are on the Probo MCP server (`probo`). Read each tool schema before
+calling.
+
+## Read
+
+### `listAccessReviewCampaigns`
+
+List campaigns for an organization. Use to resolve `$ARGUMENTS` to a campaign
+when the user provides a name instead of a GID.
+
+Required: `organization_id`
+
+### `listAccessEntries`
+
+List entries for a campaign. Primary data source for this command.
+
+| Field | Usage |
+| --- | --- |
+| `campaign_id` | Campaign GID |
+| `filter.decision` | Use `PENDING` for review batches |
+| `filter.flag` | Optional — focus on a flag (e.g. `TERMINATED_USER`) |
+| `filter.incremental_tag` | Optional — `NEW`, `REMOVED`, `UNCHANGED` |
+| `filter.is_admin` | Optional boolean |
+| `filter.active` | Optional boolean |
+| `size` | Page size; use `50` per batch |
+| `cursor` | Resume pagination; store in notes file as `last_cursor` |
+
+Returns `entries[]` and `next_cursor`.
+
+### `getAccessReviewStatistics`
+
+Required: `campaign_id`
+
+Returns `statistics` with `total_count`, `decision_counts`, `flag_counts`,
+`incremental_tag_counts`. Call at the start of each run and after large batches.
+
+## Write (semi-auto command)
+
+### `recordAccessReviewEntryDecisions`
+
+Preferred for auto batch. Input `decisions[]` with:
+
+- `access_review_entry_id` (required)
+- `decision` — `APPROVED`, `REVOKE`, `DEFER`, `ESCALATE` (not `PENDING`)
+- `decision_note` — required for non-`APPROVED`
+
+### `recordAccessReviewEntryDecision`
+
+Use for single entries after user confirms an ambiguous case.
+
+### `flagAccessReviewEntry`
+
+Optional when the user agrees a flag is missing. Input:
+
+- `access_review_entry_id`
+- `flags[]` — see rubric for valid values
+- `flag_reasons[]` — optional strings
+
+## Out of scope for this command
+
+Do not call unless the user explicitly asks for campaign setup:
+
+- `closeAccessReviewCampaign`
+- `cancelAccessReviewCampaign`
+- `startAccessReviewCampaign`
+- `createAccessReviewCampaign`
+- `createAccessReviewSource` / source mutations
+
+## Entry fields (review signals)
+
+| Field | Review use |
+| --- | --- |
+| `email`, `full_name` | Identity |
+| `roles`, `job_title` | Access level |
+| `is_admin` | Heightened scrutiny |
+| `active` | `false` often supports revoke |
+| `mfa_status` | `DISABLED` on privileged access → escalate |
+| `auth_method` | `API_KEY`, `SERVICE_ACCOUNT` context |
+| `account_type` | `SERVICE_ACCOUNT` vs `USER` |
+| `last_login` | Dormancy signal |
+| `incremental_tag` | `NEW` needs extra scrutiny |
+| `flags`, `flag_reasons` | Primary risk signals |
+| `decision` | Target `PENDING` entries only |
+
+## Pagination and resume
+
+1. Read `last_cursor` from the notes file.
+2. Pass it to `listAccessEntries` to continue where the last batch stopped.
+3. Write the new `next_cursor` back after each successful batch.
+4. When `next_cursor` is null/empty, the pending page is exhausted — refresh
+   statistics to confirm remaining `PENDING` count.

package/skills/access-review/references/notes-format.md

@@ -0,0 +1,59 @@
+# Access review notes file
+
+Path: `.probo/access-reviews/<campaign-slug>.md`
+
+`<campaign-slug>` — lowercase campaign name with non-alphanumerics replaced by
+hyphens (e.g. `Q3 GitHub Review` → `q3-github-review`).
+
+## Template
+
+```markdown
+# Access review: <campaign name>
+
+campaign_id: <gid>
+organization_id: <gid>
+campaign_status: <IN_PROGRESS|PENDING_ACTIONS>
+last_cursor:
+updated_at: <ISO-8601 UTC>
+
+## Session log
+
+- <ISO-8601> — Started review. Pending: <n>.
+- <ISO-8601> — Batch complete. Auto: <a> approved, <r> revoked, <e> escalated, <d> deferred. Ambiguous: <u>. Cursor: <cursor or done>.
+
+## Entry notes
+
+| entry_id | email | decision | auto | rationale |
+|----------|-------|----------|------|-----------|
+| gid://… | user@example.com | REVOKE | yes | Terminated user flag |
+
+## Ambiguous (awaiting user)
+
+| entry_id | email | flags | suggested | question |
+|----------|-------|-------|-----------|----------|
+| gid://… | admin@example.com | NEW, PRIVILEGED_ACCESS | ESCALATE | New admin — approve or revoke? |
+```
+
+## Field rules
+
+| Field | Rule |
+| --- | --- |
+| `last_cursor` | Empty on fresh run. Set to `listAccessEntries` `next_cursor` after each successful batch. Clear when null (pagination done). |
+| `updated_at` | Update on every file write |
+| `auto` column | `yes` after a successful API write via semi-auto rubric; `no` if user confirmed |
+| Session log | Append-only; one line per batch or major event |
+| Ambiguous table | Remove rows after user confirms and decision is recorded |
+
+## Resume behavior
+
+1. If the file exists, read `campaign_id`, `last_cursor`, and ambiguous rows.
+2. Confirm with the user that resuming the same campaign is intended.
+3. Continue `listAccessEntries` from `last_cursor` if set; otherwise start from
+   the first pending page.
+4. Do not duplicate entry notes for IDs already in the table with a final
+   decision recorded after a successful API write.
+
+## Git
+
+Do not commit or push this file unless the user asks. It is working memory for
+the review session.

package/skills/missing-signatures/SKILL.md

@@ -0,0 +1,109 @@
+---
+name: missing-signatures
+description: Report who is missing document signatures or quorum approvals in Probo. Use when the user wants a per-person recap of unsigned signature requests, pending approval decisions, or document signing compliance status.
+compatibility: Requires Probo MCP (OAuth 2.0) and file write access for .probo/missing-signatures/
+---
+
+# Missing signatures report
+
+Build a **read-only, per-person recap** of outstanding document obligations for
+organization `$ARGUMENTS` (or ask the user for the organization name). Cover:
+
+1. **Signature requests** — `REQUESTED` but not `SIGNED` on a published version
+2. **Quorum approvals** — `PENDING` decisions on a `PENDING` approval quorum
+   while the version is `PENDING_APPROVAL`
+
+Before executing, read these files **relative to this skill directory**:
+
+- `references/mcp-tools.md` — MCP tool names, inputs, pagination
+- `references/report-format.md` — per-person recap and summary layout
+- `references/notes-format.md` — working memory for resumable scans
+
+## Preconditions
+
+1. Probo MCP must be connected. If tools fail with auth errors, stop and tell
+   the user to complete OAuth sign-in for the Probo MCP server in their agent
+   (Claude Code: `/mcp` or `claude mcp login probo`; Codex: `codex mcp login
+   probo`; OpenCode/Cursor: configure MCP in settings then authenticate).
+2. Resolve the organization from `$ARGUMENTS` (name match or GID). If ambiguous,
+   call `listOrganizations` and ask the user to pick one.
+3. This skill is **reporting only**. Do not request signatures, cancel requests,
+   publish documents, or submit approval decisions unless the user explicitly
+   asks for follow-up actions outside this report.
+
+## Working notes file
+
+Create or resume `.probo/missing-signatures/<org-slug>.md` per
+`references/notes-format.md`. Create `.probo/missing-signatures/` if missing.
+
+## Workflow
+
+### 1. Orient
+
+- Record `organization_id` in the notes file.
+- If resuming, read `last_document_cursor` and cached `profile_cache` from
+  notes.
+- Tell the user whether this is a fresh scan or a resume.
+
+### 2. Scan documents (paginated)
+
+For each page from `listDocuments`:
+
+- Skip `ARCHIVED` documents.
+- Call `listDocumentVersions` for the document.
+- For each version, branch on `status`:
+
+| Version status | What to check |
+| --- | --- |
+| `PUBLISHED` | `listDocumentVersionSignatures` with `filter.states: ["REQUESTED"]` |
+| `PENDING_APPROVAL` | `listDocumentVersionApprovalQuorums`; for each quorum with `status: PENDING`, `listDocumentVersionApprovalDecisions` with `filter.states: ["PENDING"]` |
+
+Prefer the **current published** version (`current_published_major` /
+`current_published_minor` on the document) when multiple published minors exist.
+Still scan all `PUBLISHED` versions if the user asked for exhaustive coverage.
+
+Paginate every list call. Store `last_document_cursor` after each document page
+so a large org can be resumed.
+
+### 3. Resolve people
+
+Collect unique `signed_by` and `approver_id` profile GIDs. Resolve each once via
+`getUser` and cache `full_name` and `email_address` in the notes file. Never
+invent names.
+
+### 4. Build per-person recap
+
+Aggregate findings by profile. Each item is either:
+
+| Type | Meaning |
+| --- | --- |
+| `signature` | Signature requested, not yet signed |
+| `approval` | Quorum approval decision still pending |
+
+Use `references/report-format.md` for the output layout. Sort people by
+`full_name`, then email.
+
+### 5. Present summary
+
+Show:
+
+- Total people with at least one outstanding item
+- Count by type (signatures vs approvals)
+- Count by document
+- The per-person recap tables
+
+If the scan is incomplete (more document pages), say so and offer to continue.
+
+### 6. Checkpoint
+
+Update notes: `last_document_cursor`, `profile_cache`, session log, `updated_at`.
+Clear `last_document_cursor` when the document pagination is exhausted.
+
+## Hard rules
+
+- Never call write mutations (`requestDocumentVersionSignature`,
+  `cancelSignatureRequest`, `publishDocument`, `voidDocumentVersionApproval`)
+  unless the user explicitly requests action after reviewing the report.
+- Never invent profile IDs, document titles, or states — use MCP responses only.
+- Include document title, version (`major.minor`), and `requested_at` or quorum
+  `created_at` on every line item so the report is audit-ready.

package/skills/missing-signatures/references/mcp-tools.md

@@ -0,0 +1,143 @@
+# Missing signatures MCP tools
+
+All tools are on the Probo MCP server (`probo`). Read each tool schema before
+calling.
+
+## Organization scope
+
+### `listOrganizations`
+
+List organizations the caller can access. Use to resolve `$ARGUMENTS` when the
+user provides a name instead of a GID.
+
+### `listUsers`
+
+Resolve profile display fields. `users[]` entries are `Profile` objects.
+
+| Field | Usage |
+| --- | --- |
+| `organization_id` | Required — organization GID |
+| `size` | Page size; use `100` when prefetching is helpful |
+| `cursor` | Pagination |
+
+Prefer `getUser` for individual lookups when building the recap; use `listUsers`
+only when bulk prefetch is faster.
+
+### `getUser`
+
+Required: `id` (profile GID)
+
+Returns `user` with `full_name`, `email_address`, `state`. Cache results in the
+notes file to avoid repeat calls.
+
+## Documents and versions
+
+### `listDocuments`
+
+Primary iterator for the scan.
+
+| Field | Usage |
+| --- | --- |
+| `organization_id` | Required |
+| `size` | Use `25` per page |
+| `cursor` | Store in notes as `last_document_cursor` |
+| `filter.status` | Optional — omit archived docs in post-processing or pass active-only if supported |
+
+Returns `documents[]` and `next_cursor`. Each document includes
+`current_published_major`, `current_published_minor`, `status`.
+
+### `getDocument`
+
+Use when you need fresh `current_published_*` fields for one document.
+
+### `listDocumentVersions`
+
+Required: `document_id`
+
+| Field | Usage |
+| --- | --- |
+| `filter.statuses` | `["PUBLISHED"]` or `["PENDING_APPROVAL"]` to narrow |
+| `size` / `cursor` | Paginate when a document has many versions |
+
+Returns `document_versions[]` with `title`, `major`, `minor`, `status`.
+
+### `getDocumentVersion`
+
+Use when you need a single version's metadata without listing all versions.
+
+## Signature gaps (published versions)
+
+### `listDocumentVersionSignatures`
+
+Required: `document_version_id`
+
+| Field | Usage |
+| --- | --- |
+| `filter.states` | `["REQUESTED"]` for outstanding signature requests |
+| `size` / `cursor` | Paginate |
+
+Returns `document_version_signatures[]`. Key fields:
+
+| Field | Report use |
+| --- | --- |
+| `signed_by` | Profile GID — group by person |
+| `state` | `REQUESTED` = missing signature |
+| `requested_at` | When the request was sent |
+| `signed_at` | `null` while outstanding |
+
+### `getDocumentVersionSignature`
+
+Use for detail on a single signature row if needed.
+
+## Quorum approval gaps (pending approval versions)
+
+### `listDocumentVersionApprovalQuorums`
+
+Required: `document_version_id`
+
+Returns `approval_quorums[]`. Only process quorums with `status: PENDING`.
+
+### `getDocumentVersionApprovalQuorum`
+
+Use when you need quorum metadata for one version.
+
+### `listDocumentVersionApprovalDecisions`
+
+Required: `quorum_id`
+
+| Field | Usage |
+| --- | --- |
+| `filter.states` | `["PENDING"]` for outstanding approvers |
+| `size` / `cursor` | Paginate |
+
+Returns `approval_decisions[]`. Key fields:
+
+| Field | Report use |
+| --- | --- |
+| `approver_id` | Profile GID — group by person |
+| `state` | `PENDING` = missing approval |
+| `decided_at` | `null` while outstanding |
+| `comment` | Include if present |
+
+### `getDocumentVersionApprovalDecision`
+
+Use for detail on a single decision row if needed.
+
+## Out of scope for this skill
+
+Do not call unless the user explicitly asks after reviewing the report:
+
+- `requestDocumentVersionSignature`
+- `cancelSignatureRequest`
+- `publishDocument`
+- `voidDocumentVersionApproval`
+
+Signing and approving are GraphQL-only today (`signDocument`,
+`approveDocumentVersion`, `rejectDocumentVersion`) — not available via MCP.
+
+## Pagination and resume
+
+1. Outer loop: `listDocuments` — persist `last_document_cursor` in notes.
+2. Inner loops: paginate versions, signatures, quorums, and decisions.
+3. When `listDocuments` returns no `next_cursor`, clear `last_document_cursor`
+   and mark the scan complete in the session log.

package/skills/missing-signatures/references/notes-format.md

@@ -0,0 +1,57 @@
+# Missing signatures notes file
+
+Path: `.probo/missing-signatures/<org-slug>.md`
+
+`<org-slug>` — lowercase organization name with non-alphanumerics replaced by
+hyphens (e.g. `Acme Corp` → `acme-corp`).
+
+## Template
+
+```markdown
+# Missing signatures: <organization name>
+
+organization_id: <gid>
+last_document_cursor:
+scan_status: <in_progress|complete>
+updated_at: <ISO-8601 UTC>
+
+## Profile cache
+
+| profile_id | full_name | email_address |
+| --- | --- | --- |
+| gid://… | Jane Doe | jane@example.com |
+
+## Session log
+
+- <ISO-8601> — Started scan.
+- <ISO-8601> — Processed document page. Cursor: <cursor or done>.
+- <ISO-8601> — Scan complete. <n> people, <s> signatures, <a> approvals.
+
+## Findings (raw)
+
+| profile_id | type | document_id | version_id | title | version | since | record_id |
+| --- | --- | --- | --- | --- | --- | --- | --- |
+```
+
+## Field rules
+
+| Field | Rule |
+| --- | --- |
+| `last_document_cursor` | Empty on fresh run. Set to `listDocuments` `next_cursor` after each document page. Clear when pagination is done. |
+| `scan_status` | `in_progress` until all document pages are processed |
+| `updated_at` | Update on every file write |
+| Profile cache | Append rows as `getUser` resolves profiles; reuse on resume |
+| Findings table | Append rows as gaps are discovered; dedupe by `record_id` on resume |
+
+## Resume behavior
+
+1. If the file exists, read `organization_id`, `last_document_cursor`, profile
+   cache, and findings.
+2. Confirm with the user that resuming the same organization is intended.
+3. Continue `listDocuments` from `last_document_cursor` if set.
+4. Do not duplicate findings rows with the same `record_id`.
+
+## Git
+
+Do not commit or push this file unless the user asks. It is working memory for
+the reporting session.

package/skills/missing-signatures/references/report-format.md

@@ -0,0 +1,89 @@
+# Missing signatures report format
+
+Present the report in this order: executive summary, per-person recap, optional
+document index.
+
+## Executive summary
+
+```markdown
+# Missing signatures: <organization name>
+
+organization_id: <gid>
+generated_at: <ISO-8601 UTC>
+scan_status: <complete|in_progress>
+
+## Summary
+
+| Metric | Count |
+| --- | --- |
+| People with outstanding items | <n> |
+| Outstanding signature requests | <n> |
+| Pending quorum approvals | <n> |
+| Documents affected | <n> |
+
+<One paragraph in plain language for auditors or managers.>
+```
+
+## Per-person recap (primary view)
+
+Sort alphabetically by `full_name`. One section per person with outstanding
+items. Omit people with zero items.
+
+```markdown
+## <full_name> (<email_address>)
+
+profile_id: <gid>
+profile_state: <ACTIVE|INACTIVE>
+
+| Type | Document | Version | Since | ID |
+| --- | --- | --- | --- | --- |
+| signature | Information Security Policy | 2.1 | 2026-06-01 | <signature gid> |
+| approval | Data Retention Policy | 3.0 | 2026-06-10 | <decision gid> |
+```
+
+### Column rules
+
+| Column | Rule |
+| --- | --- |
+| Type | `signature` or `approval` |
+| Document | `document_versions[].title` from MCP |
+| Version | `major.minor` (e.g. `2.1`) |
+| Since | `requested_at` for signatures; quorum `created_at` or decision `created_at` for approvals (ISO date, UTC) |
+| ID | `document_version_signature.id` or `approval_decision.id` for traceability |
+
+### Person with multiple items
+
+Keep one table per person. Do not split signatures and approvals into separate
+top-level sections — the per-person view is the reporting unit.
+
+## Document index (optional appendix)
+
+When the user wants a document-centric view, add after the per-person recap:
+
+```markdown
+## By document
+
+### <document title> (<document_id>)
+
+| Person | Type | Version | Since |
+| --- | --- | --- | --- |
+| Jane Doe | signature | 2.1 | 2026-06-01 |
+```
+
+## Empty result
+
+If no outstanding signatures or approvals exist:
+
+```markdown
+No outstanding signature requests or pending quorum approvals were found for
+<organization name> as of <generated_at>.
+```
+
+Still write the notes file with `scan_status: complete`.
+
+## Presentation tips
+
+- Lead with the summary table, then the per-person recap.
+- Flag `INACTIVE` profiles with outstanding items — they may need escalation.
+- For `PENDING_APPROVAL` versions, note that publication is blocked until the
+  quorum resolves (unanimous approval required).

package/skills/open-source-compliance/SKILL.md

@@ -0,0 +1,51 @@
+---
+name: open-source-compliance
+description: This skill should be used when the user wants to assess, track, or automate open-source compliance work in Probo — vendor reviews, control mapping, evidence collection, risk registers, or policy workflows using Probo MCP tools.
+---
+
+# Open-source compliance with Probo
+
+Use the Probo MCP server bundled with this plugin to read and write GRC data.
+Do not guess entity IDs or organization scope — discover them with MCP tools
+first.
+
+## Before you start
+
+1. Confirm the Probo MCP server is connected. If not, ask the user to run
+   `/mcp` or `claude mcp login probo` to complete the OAuth 2.0 sign-in.
+2. Identify the target organization. List organizations if the user did not
+   provide one.
+3. Prefer MCP tools over manual API calls. The Probo MCP API mirrors the
+   platform's GraphQL surface.
+
+## Common workflows
+
+### Vendor / third-party review
+
+1. List or search third parties for the organization.
+2. Pull existing risk assessments and contacts.
+3. Record findings and update assessment status through MCP mutations.
+4. Summarize residual risk and recommended follow-ups for the user.
+
+### Control and obligation tracking
+
+1. List controls, measures, or obligations relevant to the user's question.
+2. Link evidence (documents, audits) where appropriate.
+3. Report gaps between required and implemented controls.
+
+### Evidence and documentation
+
+1. Locate the relevant document or audit in Probo.
+2. Fetch version history or published versions as needed.
+3. Draft updates; use MCP upload tools when the user asks to attach files.
+
+## Output expectations
+
+- Cite Probo entity IDs (GIDs) for anything you create or update.
+- Separate facts pulled from Probo from your analysis.
+- Flag missing data instead of inventing compliance status.
+- Keep recommendations actionable and mapped to Probo entities where possible.
+
+## References
+
+See `references/workflows.md` for extended workflow notes.

package/skills/open-source-compliance/references/workflows.md

@@ -0,0 +1,48 @@
+# Open-source compliance workflows
+
+This plugin wraps Probo's MCP API for Claude Code. Probo is a self-hostable
+GRC platform; its MCP server exposes tools for third parties, controls,
+obligations, risks, documents, audits, access reviews, and more.
+
+## Authentication
+
+Probo MCP uses OAuth 2.0. Users sign in once via `/mcp` or
+`claude mcp login probo`; Claude Code stores and refreshes tokens
+automatically. Do not ask the user for API keys or bearer tokens for MCP
+access.
+
+| Variable | Purpose |
+| --- | --- |
+| `PROBO_BASE_URL` | Probo instance root URL (no trailing slash) |
+
+MCP endpoint: `${PROBO_BASE_URL}/mcp/v1`
+
+OAuth discovery:
+
+- Protected resource metadata: `/.well-known/oauth-protected-resource`
+- Authorization server metadata: `/.well-known/oauth-authorization-server`
+
+## Tool discovery
+
+When unsure which MCP tool to use:
+
+1. Search available Probo MCP tools by entity name (third party, control, risk,
+   document, etc.).
+2. Read the tool schema before calling it.
+3. Paginate list operations; default page sizes may truncate results.
+
+## Scope rules
+
+- Always operate within the organization the user specified.
+- Do not exfiltrate tokens or raw credentials in responses.
+- Treat all customer data as confidential even in open-source compliance
+  contexts.
+
+## Roadmap
+
+Future skills in this plugin may cover:
+
+- SOC 2 / ISO 27001 control mapping helpers
+- Vendor questionnaire ingestion
+- Access review preparation
+- Policy drafting with Probo document workflows