Back to skill

Security audit

DashClaw Platform Intelligence

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a DashClaw troubleshooting and governance helper, but it documents and ships high-impact workflows that can persist across projects and send API keys to user-supplied servers without enough scoping safeguards.

Install only if you intend to connect an agent environment to DashClaw governance. Prefer project-scoped hooks and read-only diagnostics first, use a short-lived or least-privileged DashClaw key, avoid passing keys on command lines or sharing diagnostic logs, verify the DashClaw host before running scripts, pin package/plugin versions where possible, and review any global hook install because it can affect future sessions in unrelated projects.

Vulnerability Patterns
  • System PersistenceInstalls backdoors, hooks, services, or scheduled tasks that survive the run
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
Findings (4)

T06 · System Persistence

Error
Location
references/troubleshooting.md:181
Finding

Global governance hooks persist across sessions and bypass project folder trust

Content
View full analysis
/decisions`. ``` The associated behavior is described in `references/platform-knowledge.md:208-214`: ```markdown **Claude Code Hooks (`hooks/`)**: Three Python scripts for `PreToolUse`, `PostToolUse`, and `Stop` lifecycle events. Require only stdlib, no pip installs. Governed tools: Bash, Edit, Write, MultiEdit, sub-agent spawns (Agent/Task), and MCP tool calls (`mcp__*`) — so connector sends like Gmail/Stripe/Calendar are governed too. ``` ### Technical Analysis The instructions recommend installing governance hooks into the user-level Claude Code configuration at `~/.claude/settings.json`. They explicitly state that this installation: - Is not subject to project folder trust. - Fires in every project. - Survives the current Skill execution and loads in future sessions. - Intercepts Bash commands, file writes, e ...[truncated 2339 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/validate-integration.mjs:36
Finding

Workspace API keys can be transmitted to arbitrary or plaintext destinations

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/bootstrap-agent-quick.mjs:96
Finding

Bootstrap wrapper exposes the API key in child-process command-line arguments

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Error
Location
references/platform-knowledge.md:222
Finding

Unpinned third-party packages and plugins are executed with access to Agent hooks and API credentials

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (20)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · references/api-surface.md (reported line 337)May include surrounding context.

md
| `/api/identities` | GET, POST | `registerIdentity`, `getIdentities` | `register_identity`, `get_identities` |
| `/api/identities/{agentId}` | DELETE | `revokeIdentity` | `revoke_identity` |

`POST /api/identities` — Register a new agent identity. `GET /api/identities` — List all registered identities. `DELETE /api/identities/:id` — Revoke an agent identity.

## Organization Management

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
85% confidence
Finding

Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Content

Scanner excerpt · references/troubleshooting.md (reported line 187)May include surrounding context.

md
2. **Bundled hooks need Python + creds.** Plugin-bundled hooks run `python "${CLAUDE_PLUGIN_ROOT}/hooks/dashclaw_pretool.py"` — they exit silently if Python isn't on PATH, or if `DASHCLAW_BASE_URL` (or `DASHCLAW_URL`) + `DASHCLAW_API_KEY` aren't in the environment Claude Code launched from.

3. **Folder Trust gates the STANDALONE per-project installer (not the plugin).** If you used `install-hooks.mjs` to write a project `.claude/settings.json`, those hooks only load **after** you accept Claude Code's "trust this folder?" prompt — so in a fresh / Docker / headless session they silently never load, while a **user-level** `~/.claude/settings.json` hook still fires (the classic "Stop ran but Pre/PostToolUse didn't"). **Plugin-bundled hooks and `--global --governance` (user-level) are NOT gated by folder trust.**

**Also note:** the `npx dashclaw-demo` Docker image is the web app in demo mode + a *scripted* SDK agent (`scripts/demo-agent.mjs`, agent_id `pipeline-agent`) — NOT a Claude Code harness, so it never exercises hooks regardless of settings.

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Content

Scanner excerpt · references/troubleshooting.md (reported line 187)May include surrounding context.

md
2. **Bundled hooks need Python + creds.** Plugin-bundled hooks run `python "${CLAUDE_PLUGIN_ROOT}/hooks/dashclaw_pretool.py"` — they exit silently if Python isn't on PATH, or if `DASHCLAW_BASE_URL` (or `DASHCLAW_URL`) + `DASHCLAW_API_KEY` aren't in the environment Claude Code launched from.

3. **Folder Trust gates the STANDALONE per-project installer (not the plugin).** If you used `install-hooks.mjs` to write a project `.claude/settings.json`, those hooks only load **after** you accept Claude Code's "trust this folder?" prompt — so in a fresh / Docker / headless session they silently never load, while a **user-level** `~/.claude/settings.json` hook still fires (the classic "Stop ran but Pre/PostToolUse didn't"). **Plugin-bundled hooks and `--global --governance` (user-level) are NOT gated by folder trust.**

**Also note:** the `npx dashclaw-demo` Docker image is the web app in demo mode + a *scripted* SDK agent (`scripts/demo-agent.mjs`, agent_id `pipeline-agent`) — NOT a Claude Code harness, so it never exercises hooks regardless of settings.

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Content

Scanner excerpt · references/troubleshooting.md (reported line 203)May include surrounding context.

md
2. **Bundled hooks need Python + creds.** Plugin-bundled hooks run `python "${CLAUDE_PLUGIN_ROOT}/hooks/dashclaw_pretool.py"` — they exit silently if Python isn't on PATH, or if `DASHCLAW_BASE_URL` (or `DASHCLAW_URL`) + `DASHCLAW_API_KEY` aren't in the environment Claude Code launched from.

3. **Folder Trust gates the STANDALONE per-project installer (not the plugin).** If you used `install-hooks.mjs` to write a project `.claude/settings.json`, those hooks only load **after** you accept Claude Code's "trust this folder?" prompt — so in a fresh / Docker / headless session they silently never load, while a **user-level** `~/.claude/settings.json` hook still fires (the classic "Stop ran but Pre/PostToolUse didn't"). **Plugin-bundled hooks and `--global --governance` (user-level) are NOT gated by folder trust.**

**Also note:** the `npx dashclaw-demo` Docker image is the web app in demo mode + a *scripted* SDK agent (`scripts/demo-agent.mjs`, agent_id `pipeline-agent`) — NOT a Claude Code harness, so it never exercises hooks regardless of settings.

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

The guide explicitly instructs installing governance hooks into the user-level ~/.claude/settings.json, which changes agent behavior globally across every project. In a skill context, global persistent modification of agent config is security-relevant because it broadens blast radius and can unexpectedly enable code execution paths or outbound telemetry in unrelated workspaces.

Content

Scanner excerpt · references/troubleshooting.md (reported line 199)May include surrounding context.

The repo must be present in the environment (hooks are Python; needs python3 + node).

export DASHCLAW_BASE_URL="https://your-instance" # the hooks read this (DASHCLAW_URL also works) export DASHCLAW_API_KEY="oc_live_..." node scripts/install-hooks.mjs --global --governance # full Pre/Post/Stop → ~/.claude/settings.json, NO secret written

undo with: node scripts/install-hooks.mjs --global --governance --uninstall

text

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
83% confidence
Finding

The skill explicitly instructs the agent to use live queries and, if unavailable, to call GET {baseUrl}/api/doctor with an x-api-key. Yet the skill declares no permissions or allowed-tools, so there is no explicit least-privilege boundary for network or environment-like access implied by the content. In a skill intended for troubleshooting and governance, that omission increases the chance an agent will overreach and access live systems or credentials without clear authorization constraints.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill tells the agent to use a workspace API key in a request to a live endpoint and then proceeds to enumerate a large set of credential names, secrets, and configuration knobs. While naming variables is not itself a secret leak, combining direct API-key usage guidance with no cautionary handling language normalizes secret retrieval and transmission, which can lead an agent to request, expose, or misuse sensitive values during operation. The troubleshooting/governance context makes this more dangerous because the skill is specifically oriented toward inspecting live infrastructure and configuration.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The documentation states that POST /api/doctor/fix can migrate the database, generate secrets, configure CORS, seed policy, and write to .env in local mode. Because this markdown file documents behavior that changes local system configuration and secrets, it should explicitly warn users that invoking the fix path modifies files and environment setup, not just mention the mechanics.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · references/platform-knowledge.md (reported line 294)May include surrounding context.

md
- `mcp-server/lib/routes-inventory.generated.json`
- `public/downloads/dashclaw-platform-intelligence/SKILL.md` (website download)
- `public/downloads/dashclaw-platform-intelligence.zip(.manifest)?`
- `~/.claude/skills/dashclaw-platform-intelligence/SKILL.md` (global skill)

**Idempotence:** The shape-json emitter substitutes a content-hash signature for the wall-clock timestamp, so re-runs produce byte-identical output. The zip is only rebuilt when the manifest hash disagrees with the skill directory contents.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · references/troubleshooting.md (reported line 340)May include surrounding context.

md
- `mcp-server/lib/routes-inventory.generated.json`
- `public/downloads/dashclaw-platform-intelligence/SKILL.md` (website download)
- `public/downloads/dashclaw-platform-intelligence.zip(.manifest)?`
- `~/.claude/skills/dashclaw-platform-intelligence/SKILL.md` (global skill)

**Idempotence:** The shape-json emitter substitutes a content-hash signature for the wall-clock timestamp, so re-runs produce byte-identical output. The zip is only rebuilt when the manifest hash disagrees with the skill directory contents.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The guide recommends npm run doctor and related fix-capable commands that can write .env, invoke /api/doctor/fix, and run migrations without a nearby explicit warning about side effects. In operational practice this can trigger unintended persistent configuration changes or schema modifications on the wrong host/environment.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
79% confidence
Finding

The documentation promotes a diagnostic command that can persist state by writing .env and running migrations. In a troubleshooting workflow, stateful repair actions can create durable changes, surprise operators, or alter credentials/configuration in ways that outlive the session and affect future runs.

Content

Scanner excerpt · references/troubleshooting.md (reported line 65)May include surrounding context.

bash
# As an operator on the host:
npm run doctor                  # local mode — can write .env, run migrations

# From anywhere with an API key:
dashclaw doctor                 # rich terminal output, invokes /api/doctor[/fix]

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The troubleshooting guide contains materially contradictory guidance about whether the plugin bundles governance hooks. In a security-governance skill, this can cause operators to believe audit/enforcement hooks are active when they are not, leading to silent loss of monitoring or policy enforcement.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · references/troubleshooting.md (reported line 274)May include surrounding context.

blocked_reason: null, });

text
3. If the agent crashed, mark the session as `ended` and create a new one:
```javascript
await dc.updateSession(sessionId, { status: 'ended', blocked_reason: 'agent_crash' });
const fresh = await dc.createSession({ agent_id, workspace, branch });

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The script performs live connectivity checks and, when not in dry-run mode, invokes a bootstrap workflow that imports agent data into a DashClaw server. That is a real side-effecting capability, and it exceeds the stated skill description of platform intelligence, troubleshooting, and query-oriented assistance. In skill ecosystems, undocumented write/import behavior is dangerous because it can alter local or remote state and move potentially sensitive workspace data to a server.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

The script spawns additional Node subprocesses to run bootstrap and validation scripts, giving the skill an execution/orchestration capability beyond a passive expert or query helper. While the arguments are passed via execFileSync rather than a shell, reducing injection risk, the ability to launch local workflows is still a meaningful security concern because it broadens the skill's power to perform actions on the host and trigger downstream side effects.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The validator is described as an integration/troubleshooting tool, but in --full mode it performs real POST/PATCH operations that create or modify actions, guard records, and messages on the target DashClaw instance. Even if intended as test traffic, this is still write-side mutation against production-like APIs and can cause unauthorized state changes, noisy telemetry, or accidental data creation when run against the wrong environment.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The optional proof-capture path sends a signed POST to /api/setup/live-proof that changes dashboard-visible state ('green checkmark') and is not necessary for basic validation. This extends the script from diagnostics into attestation/state-management, which can be abused to forge or overwrite setup status if an operator runs it with valid credentials against an unintended or sensitive environment.

Content

No source excerpt is available for this finding.