The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted data such as bug reports, error logs, and captured network traces to drive the debugging process.
Ingestion points: Processes user-supplied bug descriptions and artifacts like HAR files or log dumps mentioned in Phase 1 of SKILL.md.
Boundary markers: Explicitly instructs the agent to redact secrets with <REDACTED> before processing, but lacks formal delimiters for untrusted data.
Capability inventory: The agent is authorized to run curl, CLI tools, headless browsers, git bisect, and execute generated test scripts (SKILL.md Phase 1, 4, 5).
Sanitization: Relies on manual redaction instructions rather than automated escaping or validation of ingested content.
[COMMAND_EXECUTION]: Instructions in SKILL.md Phase 1 and 5 direct the agent to execute shell commands, including curl requests, CLI invocations, and test runners to build feedback loops.
[DYNAMIC_EXECUTION]: The agent is encouraged to generate and execute reproduction scripts, unit tests, and throwaway harnesses (SKILL.md Phase 1) to isolate and fix bugs.
[EXTERNAL_DOWNLOADS]: UPSTREAM.md provides installation instructions using npx skills, which fetches the skill package from a remote repository. This is associated with the vendor's distribution mechanism.

diagnosing-bugs