The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because its core methodology involves processing untrusted external data in an environment where the agent has shell access.
Ingestion points: Instructions in SKILL.md direct the agent to ingest and analyze external data sources including network traces, HTTP payloads, event logs, HAR files, and core dumps.
Boundary markers: While the skill advises selective quoting to 'carry the signal', it does not provide explicit instructions for using delimiters or boundary markers to isolate untrusted content from the agent's instruction context.
Capability inventory: The skill maintains a broad capability set including shell command execution (curl, git), executing local scripts, and driving headless browsers (Playwright/Puppeteer).
Sanitization: The skill emphasizes the redaction of secrets and credentials but lacks specific measures to sanitize ingested data for malicious embedded instructions.
[COMMAND_EXECUTION]: The skill's primary function requires the agent to execute a variety of shell commands, test suites, and custom scripts (e.g., scripts/hitl-loop.template.sh) to reproduce and diagnose software defects. This represents a legitimate but high-privilege capability set.

diagnosing-bugs