Wurd

AdvisoryAudited by Static analysis on May 10, 2026.

Overview

Detected: suspicious.prompt_injection_instructions

Findings (1)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

NoteHigh Confidence

ASI05: Unexpected Code Execution

What this means

Running the command or loading external plugins can execute local Wurd project/plugin code while generating the HTML output.

Why it was flagged

The skill tells users how to run the Wurd compiler and optionally load plugins from a directory. Executing project/plugin code is expected for a plugin-based compiler, but it should be limited to trusted code.

Skill content

npx tsx src/cli.ts <path-to-markdown> [--no-cache] [--pdf] [--plugins <dir>]

Recommendation

Run this only in a trusted Wurd project and review external plugins before using the --plugins option.

NoteHigh Confidence

ASI03: Identity and Privilege Abuse

What this means

If enabled, the compiler can use your LLM provider account and consume quota for LLM-powered plugins.

Why it was flagged

The optional LLM-powered plugins require a provider API key and endpoint. This credential use is disclosed and tied to the stated graph/table generation features.

Skill content

LLM_API_KEY=sk-...
LLM_BASE_URL=https://api.anthropic.com/v1
LLM_MODEL=claude-opus-4-6

Recommendation

Use a dedicated or scoped API key if available, keep the .env file private, and avoid committing it to version control.

NoteHigh Confidence

ASI06: Memory and Context Poisoning

What this means

Generated LLM content may persist locally and affect later output if the cache is reused.

Why it was flagged

The skill discloses local persistence of LLM outputs. Cached generated content can remain on disk and be reused across later compilations until cleared.

Skill content

LLM responses are cached in `.cache/llm/` — use `--no-cache` to regenerate.

Recommendation

Use --no-cache or clear .cache/llm/ when working with sensitive documents or when you want fresh LLM output.

Findings (1)

warn

suspicious.prompt_injection_instructions

Location: SKILL.md:152
Finding: Prompt-injection style instruction pattern detected.