The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/contact_sheet.py executes local git commands (e.g., git rev-parse, git ls-files) via subprocess.run to manage project evidence and ensure compliance with version control ignore rules.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from web pages and CI logs.
Ingestion points: External content is read during browser inspection and task context recovery as described in SKILL.md.
Boundary markers: The 'Retrieved Evidence Boundary' section provides explicit instructions to the agent to treat such data as evidence only and disregard any instruction-like content.
Capability inventory: The skill has capabilities for local file writes to the .tigerkit/ directory and controlled subprocess execution of git in scripts/contact_sheet.py.
Sanitization: Sensitive UI fields are masked during capture by scripts/capture_overlay.js, and network request bodies are filtered according to an allowlist defined in references/safety.md.

tk-browser-verify