Back to skill

Security audit

GNO

Security checks for vulnerabilities and agentic risk

Overview

The skill coherently documents local knowledge search and related editing workflows, with explicit controls for sensitive access and automation.

Install the matching GNO executable from the named publisher and choose folders carefully. Session archives may contain sensitive conversations; keep them separate and local unless you approve sharing. Enable writes, hooks, and schedules only when wanted. For batch exports, use unique filenames and overwrite protection because the supplied example can replace earlier outputs.

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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
Findings (25)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared purpose covers document indexing, retrieval, and question answering. The supplied commands instead expose explicit fact creation and replacement in a memory collection. These are substantive memory-management capabilities, not merely supporting writes needed to build a search index, and are not disclosed in the description.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The chunk concerns managing agent skill installations rather than document indexing, retrieval, or question answering. In particular, gno skill uninstall --target all performs an undeclared removal operation unrelated to the stated search capabilities. These commands appear in README documentation; the snippet does not establish that they execute automatically.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The description covers retrieval and indexing, but the code explicitly performs a content-write operation. Capturing a new statement is a distinct, undeclared capability, not merely a supporting detail of document search. UUID generation and request-status checking support that write operation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

Policy inspection and remote-model eligibility checks can support the declared AI-search functionality. However, relaxing data-egress restrictions and deleting security audit records are material administrative capabilities not disclosed in the description. The chunk is a command reference, so it does not establish that these operations actually ran or that any data was transmitted.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The declared purpose covers retrieval, indexing, and answering questions over existing content. The supplied commands instead demonstrate a content-capture workflow for a new statement. Creating knowledge-base content is a distinct capability not explicitly declared here. Request-status checking is a supporting implementation detail, not a separate mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
89% confidence
Finding

The supplied commands perform content capture and record creation rather than searching or indexing existing documents. Although capture could complement a knowledge base, it is a distinct user-facing write capability not disclosed by the search-focused description. The chunk does not demonstrate network retrieval merely by specifying a source URL.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
89% confidence
Finding

The commands operate on a context-watch system rather than performing the declared indexing, searching, or cited question answering. Although the watch includes a document-related question, maintaining registrations and enabling notifications are additional user-facing capabilities not disclosed in the description. The chunk does not establish external data transmission or other unrelated access.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

The chunk performs a link audit rather than document indexing, retrieval, or question answering. Link auditing is a distinct capability not disclosed in the description, and the supplied chunk does not establish it as a supporting step for a declared search operation. The configuration edit appears only in comments, so it is guidance rather than an executed modification.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
87% confidence
Finding

The supplied commands expose a user-facing content-capture capability that is not stated in the description. Although captured notes could later support a knowledge base, these commands primarily add content rather than perform the declared search, directory indexing, or question-answering functions. The source URL is supplied as metadata; the snippet does not establish network access.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
86% confidence
Finding

The declared purpose covers document retrieval and question answering but does not disclose ongoing monitoring or notifications. Although related to document knowledge workflows, watch registration and notification management are distinct capabilities rather than ordinary search implementation details. The snippet does not establish how notifications are delivered or whether external services are involved.

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 569)May include surrounding context.

md
files and `--index` names for comparisons; see [cli-reference.md](cli-reference.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 640)May include surrounding context.

md
[cli-reference.md](cli-reference.md), [mcp-reference.md](mcp-reference.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 711)May include surrounding context.

md
the ID identifies one write for retries. Details: [cli-reference.md](cli-reference.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 773)May include surrounding context.

md
| Complete CLI reference (all commands, options, flags) | [cli-reference.md](cli-reference.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 775)May include surrounding context.

md
| Usage examples and patterns | [examples.md](examples.md) |

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 · cli-reference.md (reported line 953)May include surrounding context.

md
--------------------------------------------- |
| `-t, --target`   | claude-desktop | Target client (see below)                                                                                                                                              |
| `-s, --scope`    | target default | Scope: `user`, `project`                                                                                                                                               |
| `-f, --force`    | false          | Overwrite existing config                                                                                                                                              |
| `--dry-run`      | false          | Preview changes                                                                                                                                                        |
| `--enable-write` | false          | Registered server starts with write tools enabled

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% 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 · examples.md (reported line 370)May include surrounding context.

md
gno tags add gno://work/readme.md project/api

# Remove tag
gno tags rm gno://work/readme.md draft

# Tags in frontmatter are extracted automatically
cat myfile.md

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 · mcp-reference.md (reported line 65)May include surrounding context.

md
Installed entries deliberately pin the current Bun/package entrypoint, active
index, absolute config, data directory, and model cache. Standard clients use
`env`; OpenCode uses `environment`; Codex uses `~/.codex/config.toml` or project
`.codex/config.toml` with `[mcp_servers.gno]` and `[mcp_servers.gno.env]`.
Do not shorten a generated entry to `gno mcp`.

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 · SKILL.md (reported line 345)May include surrounding context.

md
Relaxing to `lan` or `remote` is a visible user-authorized write. Read the
current revision with `get`, then use
`gno collection policy set <collection> <policy> --confirm-relaxation <revision>`.
Never invent or reuse a revision. Tightening to `local_only` needs no
confirmation and invalidates active sessions, streams, and queued work.

Bearer authentication, MCP write enablement, and collection policy are

Behavior Manipulation

Medium
Category
Prompt Injection
Confidence
75% confidence
Finding

Subtle instructions detected that may alter agent decision-making or introduce hidden biases.

Content

Scanner excerpt · SKILL.md (reported line 439)May include surrounding context.

md
neighbors` for nearby documents, `gno_graph_path` for "how are X and Y connected?", `gno_links`/`gno_backlinks` for one-document link expansion, and `gno_similar` for semantic neighbors. Prefer explicit or typed edges over inferred, ambiguous, or similarity edges when confidence matters. Plain `[[Note]]` links resolve across collections of the same vault (link workspace), so backlinks and impact can name other collections; pass `collection`/`collections` to keep results inside the user's scope. Never tell users to rewrite links into `[[collection:Note]]` to make them resolve.
6. Use `gno_query_diagnose` when a known target document should have appeared but did not; it reports BM25/vector/fusion/graph/rerank stage presence and filter state.
7. Use `gno_get` with `fromLine`/`lineCount` for targeted reads, or `gno_multi_get` to batch top refs.
8. Use `gno_section` only when you need a durable section locator or must re-resolve one after edits. Prefer search → `gno_get` for ordinary retrie

The batch-processing example writes retrieved documents to basename-only paths without warning that existing files or earlier results can be overwritten.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

The batch example uses shell redirection to write each retrieved document to a basename-only destination, so documents from different directories or collections can silently overwrite one another or existing output files. Redirection also truncates the destination before retrieval succeeds, potentially destroying a previous export even when retrieval fails; this is an unsafe export example rather than evidence of malicious intent.

Content

No source excerpt is available for this finding.

The recipe creates persistent notes although the manifest describes document search and indexing, not content creation.

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

The workflow instructs the agent to run gno capture to save user content as a durable note. Creating source documents is a distinct write capability, not an incidental index or embedding update needed for the manifest's search and retrieval functions.

Content

No source excerpt is available for this finding.

The recipe creates and saves new notes, extending the manifest's document-search and indexing scope.

Medium
Category
Excessive Agency
Confidence
86% confidence
Finding

The manifest describes searching, indexing, and answering questions over local documents. These capture commands instead persist new user-authored ideas and prompt patterns, with line L48 explicitly requiring that the original idea be saved; this is content creation rather than an indexing side effect.

Content

No source excerpt is available for this finding.

The activation condition does not define which requests warrant a local-context lookup.

Low
Category
Trigger Abuse
Confidence
91% confidence
Finding

“When local context may already answer the request” could apply to almost any question or task. Without a defined request scope or exclusion conditions, the recipe may activate for unrelated requests that do not call for searching local collections.

Content

No source excerpt is available for this finding.

The activation condition is overly broad: “when accuracy, traceability, or auditability matters” can apply to almost any factual request.

Low
Category
Trigger Abuse
Confidence
91% confidence
Finding

Accuracy matters in most factual answers, so this condition does not distinguish ordinary assistance from requests requiring citation and provenance verification. The activation text does not explicitly require a request for source verification or evidence-backed citations.

Content

No source excerpt is available for this finding.