The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from user messages and external GitHub contexts (issues/PRs) which are used as inputs for multi-step AI workflows.
Ingestion points: The skill ingests user input via $ARGUMENTS and GitHub data via $CONTEXT / $ISSUE_CONTEXT, as detailed in running-workflows/running-workflows.md and authoring-workflows/variables.md.
Boundary markers: Encourages users to define "six pillars" of input (problem, desired outcome, acceptance criteria) in running-workflows/running-workflows.md to guide AI behavior, but lacks strict structural delimiters to prevent instruction hijacking from untrusted GitHub content.
Capability inventory: Workflow nodes are capable of executing arbitrary shell commands and scripts via bash: and script: nodes as defined in authoring-workflows/node-reference.md.
Sanitization: Implements a security layer where user-controlled variables are passed to scripts as environment variables rather than through text substitution, specifically to mitigate shell injection as documented in authoring-workflows/variables.md.
[COMMAND_EXECUTION]: The skill's primary purpose is to execute commands via the archon CLI. It supports bash: nodes for executing shell commands and script: nodes for running TypeScript (Bun) or Python (uv) logic.
[REMOTE_CODE_EXECUTION]: The manage-run/troubleshooting.md file provides instructions for installing Claude Code by piping a remote script from Anthropic's official domain (https://claude.ai/install.sh) into a shell. This targets a trusted organization and is a standard setup procedure for the tool.
[DYNAMIC_EXECUTION]: The Archon platform supports runtime execution of authored scripts (Python and TypeScript). Documentation in authoring-workflows/authoring-workflows.md and authoring-workflows/node-reference.md describes writing, packaging, and executing these scripts as part of the workflow lifecycle.

archon-cli