The Agent Skills Directory

[COMMAND_EXECUTION]: The skill mandates the execution of a curl command targeting http://localhost:31337/notify at the start of every workflow (Scaffold, Interview, Grill, CheckCompleteness, Reconcile, Seed, Append). This is used to provide voice and text notifications to the user about the active workflow state. Since it targets a whitelisted local endpoint, it does not pose an exfiltration risk.
[INDIRECT_PROMPT_INJECTION]: The Seed and Reconcile workflows ingest data from external sources, such as repository files (README.md, package.json, tsconfig.json), Git commit logs, and ephemeral feature files. This creates an attack surface where malicious instructions embedded in a project's documentation or code metadata could attempt to influence the agent's behavior during the ISA generation or merging process.
Ingestion points: Workflows/Seed.md reads project files and Git history; Workflows/Reconcile.md reads ephemeral Markdown files.
Boundary markers: The skill relies on structured Markdown headers and YAML frontmatter to separate instructions from data.
Capability inventory: The agent has capabilities to execute shell commands (curl), read/write files in the ~/.claude/LIFEOS/ directory, and interact with Git.
Sanitization: The workflows include strict validation steps for entry types (Decisions, Learning, Verification) and enforce ISC ID stability to prevent structural corruption of the artifact.

ISA