The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow that analyzes external codebases and documentation to plan and execute code changes. This creates a surface for indirect prompt injection if the analyzed project files or remote documentation contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Processes user-supplied task descriptions ($ARGUMENTS) and scans the local project directory and external documentation during the 'Analyze' phase (references/01-analyze-code.md).
Boundary markers: The workflow is highly segmented, but there are no specific prompt delimiters or instructions to ignore embedded commands within the analyzed data.
Capability inventory: The skill utilizes shell command execution (via git, jq, and various build tools) and file system modification tools (Edit, Write).
Sanitization: The skill correctly uses jq --arg to escape user-provided task descriptions when storing them in the .claude/apex/task.json state file, mitigating basic shell injection risks.
[COMMAND_EXECUTION]: The skill frequently instructs the agent to execute shell commands for git operations, state management (using jq), and project building/linting (e.g., npm, composer, xcodebuild). These are standard development tools, but they represent significant local capabilities.
[EXTERNAL_DOWNLOADS]: The workflow includes steps to install dependencies via standard package managers like npm, pnpm, bun, and composer. These operations target well-known and trusted package registries.

apex-methodology