The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references and fetches process guidelines, engineering principles, and composition rules from the author's GitHub repository (github.com/csark0812/toolbox). These are static Markdown files used to guide the agent's behavior.
[COMMAND_EXECUTION]: The skill's workflow involves executing a local utility script (scripts/log.mjs) via Node.js to maintain a structured decision log. Additionally, it requires the agent to run a 'measurement harness' (such as a benchmarking command) to evaluate the performance of code changes.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project-specific workloads and benchmarks ('harnesses') during its optimization loop, which creates a surface for indirect instructions.
Ingestion points: Project-provided performance workloads, test suites, and benchmarking scripts used to reproduce issues.
Boundary markers: The skill does not explicitly define delimiters to separate workload output from agent instructions.
Capability inventory: Execution of shell commands (logging script and harness), file system modifications via Git (staging changes, committing, and reverting), and reading/writing log files.
Sanitization: The included logging script (log.mjs) implements sanitization to escape characters that could trigger spreadsheet formula execution (CSV/TSV injection).

hillclimb