The Agent Skills Directory

[Command Execution] (HIGH): The 'scripts/benchmark.sh' script is vulnerable to flag injection. It constructs a command string using the 'BENCH_PATTERN' variable without quoting and then executes it unquoted. This allows an attacker to inject the '-exec' flag into the 'go test' command, enabling arbitrary command execution. Evidence: 'BENCH_CMD="go test -bench=$BENCH_PATTERN ..."' and '$BENCH_CMD' in 'scripts/benchmark.sh'.\n- [Unverifiable Dependencies & Remote Code Execution] (MEDIUM): The 'scripts/lint.sh' script executes 'go mod tidy', which downloads external Go modules from the internet. This can lead to remote code execution if a malicious or compromised module is fetched. Evidence: 'go mod tidy' in 'scripts/lint.sh'.\n- [Dynamic Execution] (MEDIUM): The 'benchmark.sh' script assembles a command string at runtime and executes it. This practice is inherently risky and, combined with unquoted variables, leads to the flag injection vulnerability mentioned above. Evidence: Variable assembly of 'BENCH_CMD' in 'scripts/benchmark.sh'.\n- [Indirect Prompt Injection] (LOW): The skill is vulnerable to Indirect Prompt Injection (Category 8). It processes user-provided Go projects and command-line arguments without sanitization, providing a surface for malicious inputs to influence the execution of Go tools. 1. Ingestion points: 'project-dir' and 'pattern' arguments in 'scripts/lint.sh' and 'scripts/benchmark.sh'. 2. Boundary markers: None identified. 3. Capability inventory: Command execution via 'go test' flag injection and module downloads via 'go mod tidy'. 4. Sanitization: None.

golang-master