The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill identifies the need for and suggests the installation of the octocode-mcp NPM package if the required MCP tools are not found in the environment.
[REMOTE_CODE_EXECUTION]: In the discovery phase, the skill provides a configuration snippet that uses npx -y octocode-mcp to fetch and execute the Octocode MCP server from a remote registry.
[COMMAND_EXECUTION]: The skill uses local shell-level tools (MCP tools) and git diff to acquire code targets, search for patterns, and perform file modifications during the 'Resurrection' phase.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes untrusted code content from the local environment.
Ingestion points: Code content is read using localGetFileContent and searched via localSearchCode (referenced in SKILL.md).
Boundary markers: The instructions do not define strict delimiters to isolate code content from instructions for the agent, though it does follow a specific analysis structure.
Capability inventory: The agent has the ability to execute shell-level search tools and modify local files (Phases 1-6).
Sanitization: The skill implements 'Law 3' to redact the values of hardcoded secrets, but lacks general sanitization to prevent the agent from obeying instructions embedded in code comments.

octocode-roast