wiki-self-heal
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPERSISTENCE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill researches the web and audits local files to automatically generate new wiki content. This creates a surface where malicious instructions embedded in web pages or local wiki files could influence the agent's behavior during the 'heal' loop.
- Ingestion points: Web research results (via tools like Exa or WebSearch) and local markdown files in the 'wiki/' directory.
- Boundary markers: Absent. There are no instructions to use delimiters or ignore embedded commands when processing content from these sources.
- Capability inventory: The skill allows for file creation/modification, execution of shell commands (git, grep, find), and further network operations.
- Sanitization: Absent. While the skill defines quality gates for source credibility, it does not mandate sanitizing the retrieved text for prompt injection patterns.
- [COMMAND_EXECUTION]: The skill provides and utilizes shell command templates (using 'grep', 'find', 'git', and 'wc') to perform the wiki audit. While these are for legitimate auditing purposes, they represent a capability that could be targeted if inputs (like filenames or wiki link text) are maliciously crafted.
- [EXTERNAL_DOWNLOADS]: The skill is designed to fetch information from the internet using research tools (Exa, Firecrawl, Apify, WebSearch, WebFetch) to fill knowledge gaps.
- [PERSISTENCE]: The documentation includes templates for setting up automated schedules using macOS 'launchd', 'cron', and GitHub Actions. While these are user-implemented recipes, they facilitate the long-term autonomous execution of the skill.
Audit Metadata