The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local scripts to perform the necessary system operations for session handoff.
The agent is instructed to run node scripts/gather-state.mjs to collect git metadata including branch names, repository status, diffs, and recent commit logs.
The agent uses bash scripts/copy-clipboard.sh to interact with the system clipboard via platform-specific tools like pbcopy, wl-copy, xclip, or clip.exe to store the generated resume prompt.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to create instructions for future AI sessions, which establishes a pathway for potentially malicious content to persist across session resets.
Ingestion points: Future agent sessions ingest the "resume prompt" (via clipboard paste or an optional auto-load hook) and are instructed to read the project narrative stored at ~/.craftkit/handoff/docs/.
Boundary markers: The skill uses structural XML-style tags (<context>, <task>, <rules>) to delimit different sections of the generated handoff artifacts.
Capability inventory: The skill has the capability to write files to the local filesystem (~/.craftkit/handoff/) and modify the system clipboard. The subsequent session consuming these artifacts may have various tool-based capabilities.
Sanitization: While the skill documentation mandates the redaction of secrets, tokens, and personal data, it does not provide explicit mechanisms to sanitize or escape instructional text that might influence the behavior of the next session's agent.

craft-handoff