The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted job descriptions and search results from external websites, which could contain hidden instructions designed to manipulate the AI during fit assessment or job application tasks. 1. Ingestion points: Content retrieved via WebFetch from job URLs and snippets from WebSearch (Step 2 and Step 1c). 2. Boundary markers: None. The instructions do not implement delimiters or warnings to isolate untrusted web content from the agent's logic. 3. Capability inventory: The agent can execute shell commands via Bash, write to the local file system (seen_jobs.json, job_search_tracker.csv), and call other tools via the Agent tool. 4. Sanitization: No sanitization, filtering, or validation of the fetched web content is performed before it is processed by the model.
[REMOTE_CODE_EXECUTION]: The skill uses a dynamic execution pattern to find and run local TypeScript files using bun from computed file paths. Evidence: The skill is configured to run bun run .agents/skills/*/cli/src/cli.ts *, discovering these scripts by scanning the .agents/skills/ directory for SKILL.md files at runtime.
[COMMAND_EXECUTION]: The skill relies on shell command execution via the Bash tool to perform system checks and interface with portal-specific command-line tools.

scrape