The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from bug reports, logs, and screen recordings (videos/images) to drive its investigation and multi-engine analysis. This creates a surface where malicious instructions embedded in the analyzed data could potentially influence the agent's behavior.
Ingestion Points: Processes bug reports, logs, traces, and video files (MP4/MOV/WebM) as defined in SKILL.md and reference/video-bug-analysis.md.
Boundary Markers: Utilizes structured output schemas (e.g., video-bug-detection.schema.json) to constrain the output of analysis subagents.
Capability Inventory: The skill calls subagents (Codex, Antigravity, Claude) and diagnostic CLI tools based on the content of the investigation.
Sanitization: Employs PII masking and redaction during the video preprocessing stage as described in reference/video-bug-analysis.md to mitigate data exposure risks.
[COMMAND_EXECUTION]: The skill orchestrates the execution of numerous system-level diagnostic utilities and profilers to identify root causes of software failures.
Evidence: The skill provides workflows for running tools such as py-spy, clinic flame, go tool pprof, async-profiler, memray, and jmap as seen in the perf and memory recipes.
Evidence: The video recipe involves executing the codex exec CLI tool to perform visual analysis on extracted video frames, as detailed in reference/video-bug-analysis.md.

scout