The Agent Skills Directory

[COMMAND_EXECUTION]: Several scripts, including coordination.py, project_state.py, and retire_worktree.py, utilize the subprocess module to execute git commands and other system utilities. These operations are fundamental to the skill's purpose of managing project state, worktrees, and repository metadata.
[DYNAMIC_EXECUTION]: The retire_worktree.py script stages a reconciler script (checkpoint_sync.py) to a temporary location and executes it via sys.executable. This mechanism is designed to ensure cleanup and reconciliation logic can complete reliably even if the source repository is restructured. The codex_dispatch.py utility also executes the codex binary to perform non-interactive model operations.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from a shared coordination board (active-sessions.md) which records messages and status updates from multiple agent sessions. While this presents an injection surface, the risk is mitigated by the skill's use of structured Markdown parsing and a dedicated security utility (peer_send_gate.py) that monitors and validates message routing intents.
[OBFUSCATION]: A BIP-39 word list used for speakable session identifiers is stored in a compressed and Base85-encoded format (session-words-v1.txt.zlib.b85). This is a legitimate implementation detail for handling static assets and does not represent a malicious attempt to hide executable code.

synthesis-project-management