The Agent Skills Directory

[PERSISTENCE]: The skill includes functionality in scripts/_lib/doctor.mjs to install hooks into the Claude Code configuration file (.claude/settings.local.json). These hooks allow the skill's logic to execute automatically during SessionStart, UserPromptSubmit, and Stop events, which is the intended mechanism for platform integration.
[COMMAND_EXECUTION]: Multiple scripts, including scripts/relay.mjs, scripts/_lib/execute.mjs, and scripts/_lib/probe.mjs, utilize child_process.spawn to interact with system tools such as git, node, and uv. The skill implements safe quoting for shell arguments on Windows in scripts/_lib/doctor.mjs to mitigate command injection risks.
[DYNAMIC_EXECUTION]: In scripts/memory.mjs, the skill dynamically generates and executes a Python script via python -c to interface with the mnemosyne-memory library. This bridge is hardcoded and used for local data storage.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from handoff.json and the local workspace state. It mitigates injection risks through a mandatory evidence chain: it performs strict JSON schema validation, enforces character limits (240 chars) on strings, uses SHA256 digests for integrity verification, and provides clear boundaries via its structured checkpoint format.
[DYNAMIC_CONTEXT_INJECTION]: The hook-adapter.mjs script executes relay.mjs inject during the SessionStart event to provide contextual information to the agent about pending handoffs. The output is bounded and sanitized before being injected into the agent's context.

relay