The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/planctl.rb utility executes shell commands defined in the plan/manifest.yaml configuration file using the Open3.popen2e method. Since these commands are executed within a shell environment (sh -lc), there is a risk of arbitrary command execution if the AI agent is manipulated into inserting malicious commands into the manifest during the generation process.
[DYNAMIC_EXECUTION]: The primary workflow of the skill involves the AI agent dynamically generating executable check commands (e.g., build and test scripts) in the plan/manifest.yaml file, which are subsequently executed by the planctl script at runtime.
[DATA_EXFILTRATION]: The planctl complete command automatically stages all workspace changes using git add -A and pushes them to the configured remote repository. This process can lead to the unintentional exfiltration of sensitive files, credentials, or environment variables to the remote server if they are not explicitly excluded by a .gitignore file.
[INDIRECT_PROMPT_INJECTION]: The skill ingests project descriptions via the argument-hint parameter and uses them to generate the plan/manifest.yaml configuration. The shell commands generated in this manifest are executed by the planctl script without sanitization or boundary markers, creating an attack surface where malicious input could influence the agent to generate and run unauthorized commands.

phase-contract-workflow