Back to skill

Security audit

Tokenrip CLI

Security checks for vulnerabilities and agentic risk

Overview

The skill’s workspace, sharing, and API capabilities are disclosed and supported by clear user controls.

Install if you intend the agent to manage your Tokenrip account and workspace content. Restrict provider connections to necessary endpoints and users, keep automatic task permissions narrow, and review hosted skills before installing their persistent stubs. Link sharing exposes retained versions too; row deletion and version pruning are permanent. Prefer reviewed package versions. Scanner coverage gaps were independently investigated, but the external CLI and server implementations were not included for verification.

Vulnerability Patterns
  • 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
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (19)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The supplied commands explicitly install skills into local agent directories. Local skill installation is not declared, and the description expressly excludes local-file use. This is a distinct local-environment modification, not merely an implementation detail of publishing workspace content. The chunk does not establish what the installed skills contain or whether they subsequently execute.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

Bulk-moving artifacts aligns with the declared folder-organization capability. Bulk deletion is a distinct destructive operation, even though it is temporarily reversible, and is not covered by the description's stated publishing, organizing, or collaboration functions.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
86% confidence
Finding

Workspace inspection and membership management reasonably support the declared workspace and sharing functions. However, destructive workspace deletion and trash restoration are additional lifecycle-management capabilities not disclosed by the description. The documented administrator and approval requirements constrain deletion but do not make that capability declared. This chunk documents commands rather than executing them.

Content

No source excerpt is available for this finding.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 128)May include surrounding context.

md
rip task start --workspace-id <ws> --type <t> --step <s>` claims the next one (`references/tasks.md`) |

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 241)May include surrounding context.

md
okenrip never interprets them; your instructions say what each step means. See `references/tasks.md`.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 452)May include surrounding context.

md
A writer agent and a publisher agent pass one task along by its tags: see `references/tasks.md` § Worked example.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 458)May include surrounding context.

md
- `references/tasks.md`: type and step, start by tags and the claim loop; proposing (text and files), review rounds, send back and cancel; reporting; instructio

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 502)May include surrounding context.

md
e task is cancelled or done: `rip task reopen` it if the work is not finished (`references/tasks.md` § Error recovery) |

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 139)May include surrounding context.

md
nd the schema. The server answers `CONFLICT` if a guard you sent is stale. See `references/workspaces.md`.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 226)May include surrounding context.

md
fused call writes no row. Agents sharing a row: a plain append has one winner (`references/workspaces.md` § Coordinating agents on a row). `--strict` refuses un

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 401)May include surrounding context.

md
only when someone else wrote it, and append and `--find`/`--replace` never do (`references/workspaces.md` § Content inside a workspace). A `patch` of `--descrip

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 457)May include surrounding context.

md
- `references/workspaces.md`: load and its order, instructions, agents' handles, why, sessions, pins, browser context and navigation, adopt, internal/shared aud

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 483)May include surrounding context.

md
add or change a passage, append or use `--find`/`--replace`, which need none (`references/workspaces.md`) |

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 297)May include surrounding context.

md
A **skill** is a versioned folder of instructions for a recurring job (`SKILL.md` with `name` and `description` frontmatter, plus supporting files), owned by yo

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · references/skills.md (reported line 80)May include surrounding context.

md
hidden file or folder (such as `.env`) stops the publish with `INVALID_ARGS`: move it out of the folder. Nothing hidden is uploaded silently.
- Files that differ from the base version are sent, and a file of the base version that is no longer in the folder is removed.
- The first publish creates the skill, private. Each later publish adds a version; agents always get the latest.
- The base version is sent as the expected version. If anyone published after it, nothing is published and the server answers `CONFLICT` with the current version: run `rip skill get <name> --dir <new-folder>`, merge your changes into the new folder, and publish that folder. Their changes are never reverted silently. A folder that is not a checkout is compared with the current version, so it can undo a teammate's changes if it is older than what is on the server; edit skills in a folder fetched with `--dir`.

To edit a skill: `rip skill get <name> --dir ./<name>`, change the files, `rip skill publish ./<name>`.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
90% confidence
Finding

The update procedure invokes npx skills without a version, allowing a mutable npm release to execute code with the agent process's permissions. This is an unpinned package-execution risk rather than an MCP server issue; requiring the user's approval before updating reduces exposure but does not establish package integrity.

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 · SKILL.md (reported line 519)May include surrounding context.

md
Then refresh this skill file the way it was installed:
- **Claude Code / skills CLI:** `npx skills add tokenrip/cli`
- **ClawHub / OpenClaw:** `clawhub update tokenrip-cli`
- **Claude Cowork or by hand:** copy https://tokenrip.com/.well-known/skills/tokenrip/SKILL.md and its `references/`

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/connections.md (reported line 73)May include surrounding context.

md
# Team-owned, custom auth header, static default header
export MINIMAX_KEY=sk-...
rip connection create --team quintel --name minimax \
  --base-url https://api.minimax.io/anthropic --auth-type header \
  --auth-header-name x-api-key --secret-env MINIMAX_KEY \
  --allowed-paths 'POST /v1/messages' --header anthropic-version=2023-06-01

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · references/workspaces.md (reported line 238)May include surrounding context.

md
yer's plan and usage ("the owner's account" or "the team"), never trimmed. When a notice follows it, tell your person; don't delete anything on your own
- `Actions:` (internal members) — the payer's actions this month against its plan and when they reset, never trimmed. When a notice follows it, tell your person
- `Instructions (v<n>)`: what your person wrote for this agent in this workspace, **whole and never trimmed**, then `On its own (task types):` the types of proposal your person lets run without asking (applied for you by `rip task propose`). Without any: "No instructions here yet: your person writes them in the dashboard, Members & settings". Only your person edits them, in the dashboard; no agent writes instructions, and a change reaches you on your next load
- `Waiting on you`: up to 20 tasks for exactly this agent, in this order: `carry out` (approved: `rip task show <id>` shows the approved text and files and who approved them; carry out only that), `decide` (a review addre