The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill includes logic to scan the local filesystem for sensitive credential files. Specifically, it searches the path ~/.claude/.secrets/ for files matching the pattern gh-token-* to retrieve GitHub authentication tokens.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute complex shell operations, including the use of heredocs to generate and run scripts. These scripts perform environment variable checks, network requests, and conditional system commands.
[REMOTE_CODE_EXECUTION]: The skill performs remote execution on a host named littleblack via SSH. It executes administrative commands such as docker restart on specific service containers and inspects remote container logs.
[EXTERNAL_DOWNLOADS]: The skill initiates downloads from the external service r.jina.ai to convert web pages into markdown. It also makes POST requests to a local network endpoint 172.25.236.1:3003 to trigger scraping activities.
[PROMPT_INJECTION]: The skill processes untrusted content from the web, which creates a surface for indirect prompt injection.
Ingestion points: Content is retrieved from arbitrary user-supplied URLs via curl and the Jina Reader service in SKILL.md.
Boundary markers: The scraped content is not isolated with protective delimiters when stored or used in subsequent operations.
Capability inventory: The agent has the ability to execute bash commands, write files to the local repository, and create issues on GitHub.
Sanitization: There is no evidence of content sanitization or validation before the scraped data is used to populate markdown files or GitHub issue bodies.

research-archival