The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The agent's workflow relies on project-specific files such as KWCODE.md, PROJECT.md, and EXPERT.md to provide context and rules for task execution. These files represent an attack surface where malicious instructions could be embedded in data processed by the agent.
Ingestion points: Identified project metadata files KWCODE.md, EXPERT.md, PATTERN.md, and REFLECTION.md in SKILL.md.
Boundary markers: The documentation does not specify explicit sanitization or isolation markers for content ingested from project files.
Capability inventory: The skill possesses the ability to modify the local file system, execute testing commands (e.g., pytest), and use runtime tracing tools (sys.settrace).
Sanitization: Not explicitly documented for project-level context files.
[DYNAMIC_EXECUTION]: The agent incorporates a "Debugger" stage that utilizes sys.settrace to monitor runtime variable values and a "Verifier" stage that executes pytest against generated code. This involves the execution of code within the local project environment to validate generated patches.
[COMMAND_EXECUTION]: The skill instructions facilitate the execution of shell commands for package installation, environment setup, and running external tools like Docker for search indexing.
[EXTERNAL_DOWNLOADS]: The installation process references external resources, including the Python Package Index and a mirror provided by Tsinghua University, to fetch the kwcode package and its dependencies.

kwcode-local-coding-agent