The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a bash script scripts/run_local_valgrind.sh to coordinate a measurement loop involving docker build, docker run, and pnpm run. Inside the containerized environment, it performs runtime compilation of Rust code using cargo build and executes the resulting binaries through valgrind with the callgrind tool.
[PRIVILEGE_ESCALATION]: The Docker execution environment is configured with --cap-add SYS_PTRACE and --security-opt seccomp=unconfined. These settings grant the containerized process the ability to monitor and intercept system calls of other processes and bypass standard security profiles. While necessary for Valgrind's instrumentation to work, this represents a reduction in the security isolation between the container and the host system.
[PRIVILEGE_ESCALATION]: Upon completion of the measurement, the script executes chmod -R a+rwX on the /results directory inside the container. Because this directory is a volume mount from the host, it changes the permissions of the local output files to be world-readable, world-writable, and executable.
[INDIRECT_PROMPT_INJECTION]: The skill contains a vulnerability surface where it ingests untrusted data such as user-supplied benchmark filters and benchmark output logs.
Ingestion points: User-provided bench_filter parameter; stdout/stderr logs from benchmark binaries; Callgrind profile files.
Boundary markers: None explicitly implemented to separate instructions from processed data.
Capability inventory: Local shell execution via bash; Docker container management; Git operations (commit/push); file system writes.
Sanitization: The run_local_valgrind.sh script performs basic validation on the bench_target and repeat_count parameters, but does not sanitize the benchmark filter or the contents of logs used for reporting.

rspack-perf-valgrind-goal