The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the specweave command-line utility with multiple flags (--reason, --summary, --next, etc.) to generate handoff reports. These flags incorporate context provided by the agent or user.
[DATA_EXFILTRATION]: The skill is designed to aggregate project-wide context, including uncommitted git diffs, local configuration files (config.json), and planning documents (plan.md). While it includes heuristic secret scrubbing, the resulting artifact centralizes sensitive development data for transfer to external environments or tools.
[PROMPT_INJECTION]: The instructions include directives such as "Surface the CLI output verbatim, in order" and "do not reorder or paraphrase", which are designed to override the agent's natural language generation behavior in favor of exact output replication.
[PROMPT_INJECTION]: The skill creates an indirect attack surface by generating a document intended to be processed by other AI agents.
Ingestion points: Reads from plan.md, config.json, and git command outputs.
Boundary markers: No explicit boundary markers or "ignore embedded instructions" warnings are identified to protect the receiving agent from instructions embedded in the source data.
Capability inventory: The skill uses subprocess calls to execute the specweave CLI tool.
Sanitization: The skill uses heuristic secret scrubbing for sensitive data, but does not provide specific sanitization or escaping for potential prompt injection content within the aggregated files.

sw/handoff