The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's setup documentation in 'setup/setup-and-config.md' directs users to install the Archon CLI using a 'curl | bash' command from the external domain 'https://archon.diy/install'. This practice executes remote code without prior local verification of the script content.
[CREDENTIALS_UNSAFE]: Instructions in 'setup/setup-and-config.md' encourage users to add sensitive files, specifically '.env' files, to the 'worktree.copyFiles' configuration. This results in the exposure of local secrets, API keys, and other credentials to the AI agent during the execution of repository-based workflows.
[COMMAND_EXECUTION]: The skill provides an environment for defining and running workflows that include arbitrary shell commands ('bash:' nodes) and scripts ('script:' nodes using Python/uv or TypeScript/bun).
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input from user messages and repository content. Mandatory Evidence Chain: 1) Ingestion points: User input via '$ARGUMENTS' and file data from repository worktrees. 2) Boundary markers: Recommends using rubrics in verifier prompts but lacks mandatory system-enforced delimiters for all inputs. 3) Capability inventory: Full execution of shell commands and scripts, git repository manipulation, and file system access. 4) Sanitization: Employs environment variables for data passing but utilizes shell-quoted string splicing for '$LOOP_USER_INPUT' in loop_group bash nodes, creating a potential command injection surface.
[DYNAMIC_EXECUTION]: The core functionality relies on runtime execution of YAML-defined workflows and includes an 'exec-code' flag for test fixtures in 'authoring/variables.md', allowing the execution of arbitrary code during validation steps.

archon-cli