The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes an internal shell script located at ${CLAUDE_PLUGIN_ROOT}/scripts/larch.sh using the Bash tool. The invocation is controlled through specific subcommands and flags that are validated against a defined enum (design, implement, debate) before execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes larch run logs and token report JSON files to generate its analysis. This constitutes a data ingestion surface where external content is incorporated into the agent's context.
Ingestion points: Reads synchronized logs and unpacked cache files from the current Git repository and session-specific cache paths.
Boundary markers: No specific delimiters or boundary warnings are defined in the instructions for the log ingestion phase.
Capability inventory: The skill utilizes Read for file access and Bash for executing the larch.sh reporting utility.
Sanitization: Input validation is performed on the --skill parameter to ensure only permitted values are passed to the underlying CLI tool.

report-tokens