The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs local command execution to verify environment connectivity with its dependencies.
Evidence: The tau-doctor command in scripts/create_status_surface.py invokes a sibling skill's entrypoint (tau/run.sh) using subprocess.run with a list-based command argument to prevent shell injection.
The execution is non-mutating and is restricted to local diagnostic checks within the vendor's own ecosystem.
[PROMPT_INJECTION]: The skill manages a potential indirect prompt injection surface by ingesting machine-readable progress data and rendering it into human-readable formats.
Ingestion points: Data is read from files specified by the --input flag in scripts/create_status_surface.py.
Boundary markers: The skill automatically generates comprehensive claims and fail_closed_on metadata in its outputs to clearly define its scope and prevent agents from assuming unauthorized capabilities.
Capability inventory: The skill is restricted to reading/writing local files and executing the local tau wrapper script.
Sanitization: Input labels and IDs are escaped using html.escape before being rendered into the HTML status surface to mitigate cross-site scripting risks and unintended instruction parsing.
[DATA_EXPOSURE]: Analysis confirms the skill only accesses user-provided file paths and the local /tmp directory for report generation.
No hardcoded credentials, sensitive system paths, or network exfiltration patterns were detected in the source code or build scripts.

create-status-surface