The Agent Skills Directory

[COMMAND_EXECUTION]: The plugin template in templates/pwf-enforcer.ts utilizes node:child_process.execSync to run local scripts (inject-plan.sh and check-complete.sh). These executions are scoped to the planning-with-files dependency and are used to retrieve plan status.
[PROMPT_INJECTION]: The skill intentionally uses prompt injection as a core feature. It leverages OpenCode hooks such as experimental.chat.system.transform and experimental.session.compacting to inject planning reminders and ensure the plan remains in the agent's context during long sessions.
[EXTERNAL_DOWNLOADS]: The verification instructions in SKILL.md suggest the installation of standard development packages @opencode-ai/plugin and typescript via npm to perform syntax checks on the plugin file.
[SAFE]: The skill demonstrates safe practices, such as using JSON.stringify and shell quoting when building commands for bash injection. No evidence of data exfiltration, obfuscation, or malicious persistence was found.

pwf-enforcer