The Agent Skills Directory

[NO_CODE]: The skill is composed entirely of Markdown documentation and YAML configuration. It does not contain any executable scripts (Python, JavaScript, shell) or binary files.
[SAFE]: No malicious patterns such as prompt injection, data exfiltration, or obfuscation were detected. The playbooks emphasize evidence-driven work and safety boundaries, directing the agent to seek user authorization for irreversible actions, material costs, or handling sensitive data (as seen in references/autonomy-and-delegation.md).
[INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow for processing external data (repository code and user requests), which represents a standard attack surface. However, the instructions provide strong mitigations through structured verification steps.
Ingestion points: The agent ingests the "request and repository context" as defined in the Frame step of SKILL.md.
Boundary markers: The workflow requires the agent to explicitly define a "finish line" and an "authority boundary" before performing work.
Capability inventory: The instructions mention running "safe commands," "instrumenting state," and "building probes" (in references/autonomy-and-delegation.md and references/evidence.md), implying the use of the agent's host environment tools.
Sanitization: The skill mandates fresh evidence and manual verification of all claims and diffs before completion.

rigor