The Agent Skills Directory

[PROMPT_INJECTION]: No malicious instruction overrides or safety bypass attempts were detected. The skill uses instructional language appropriate for its debugging context. The skill ingests untrusted data (logs, user reports), which constitutes an indirect prompt injection surface.
Ingestion points: User bug reports, external artifacts (logs, HAR files, core dumps), and responses to interactive shell script prompts in scripts/hitl-loop.template.sh.
Boundary markers: Absent; the instructions do not specify markers to delimit or ignore instructions within these external sources.
Capability inventory: High; the skill instructs the agent to execute shell commands, build test harnesses, perform network requests, and modify codebase files.
Sanitization: None described; the agent is expected to process the raw content of logs and artifacts directly.
[COMMAND_EXECUTION]: The skill facilitates the creation and execution of debugging scripts and reproduction loops (e.g., tests, curl scripts, CLI tools). It includes a template script (scripts/hitl-loop.template.sh) for interactive debugging. These actions are within the expected scope of a software diagnosis tool.
[EXTERNAL_DOWNLOADS]: The skill mentions using standard utilities like curl or browser automation tools for reproduction. It does not contain instructions to download or execute code from unknown remote sources.
[DATA_EXFILTRATION]: No unauthorized data transfer patterns were found. The skill suggests analyzing logs and traces provided by the user, which is a standard debugging requirement.
[CREDENTIALS_UNSAFE]: No hardcoded secrets, API keys, or unsafe credential management practices were identified.

diagnose