The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses node:child_process to execute local git commands for capturing workspace snapshots and identifying project roots. It also generates and configures shell commands for lifecycle hooks that are executed by AI agent hosts.
[PERSISTENCE]: The skill modifies project-local and user-scope configuration files (such as ~/.claude/settings.json or .codex/hooks.json) to install lifecycle hooks. These hooks ensure the continuity runtime is invoked during host events like compaction or session startup. This behavior is documented and requires explicit user opt-in via the --hosts parameter.
[INDIRECT_PROMPT_INJECTION]: The skill represents an attack surface for indirect prompt injection as it ingests Markdown content from Work-Item Capsules and legacy checkpoints to summarize them for the agent. However, it mitigates risks by implementing a redactSensitive function to mask credentials and using structured boundary headers for the injected context.
[DYNAMIC_EXECUTION]: The skill dynamically generates POSIX and PowerShell command strings to be used as lifecycle hooks. It uses Base64 encoding for PowerShell commands to ensure cross-platform robustness and avoid shell quoting issues. The source code for these generated commands is hardcoded and scoped to calling the skill's own runtime script.
[SAFE]: The skill implements significant security best practices, including a path resolution utility (resolveWithin) that prevents directory traversal by ensuring all file operations stay within the project scope. It also features a comprehensive redaction system that masks API keys, private keys, JWTs, and authorization headers before they are persisted or shared with an agent.

setup-universal-agent-skills