Skills
Navigate
skills/dobroslavradosavljevic/skills/self-improve/Gen Agent Trust Hub

self-improve

Pass

Audited by Gen Agent Trust Hub on Oct 10, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves analyzing potentially untrusted data from the repository to influence future agent instructions and generated code.\n
  • Ingestion points: The skill reads repository-specific files (e.g., README.md, package.json), code comments, and git history as described in SKILL.md and references/harvest.md to identify workflows.\n
  • Boundary markers: The skill uses <!-- self-improve:start --> and <!-- self-improve:end --> tags in AGENTS.md to scope the generated instructions.\n
  • Capability inventory: The skill possesses the ability to write to the file system (scripts/, runbooks/, AGENTS.md) and execute commands (running bun scripts and repository toolchains).\n
  • Sanitization: A critical user approval step (Step 5 in SKILL.md) requires the user to review all discovered candidates and planned code fixes before any implementation occurs.\n- [COMMAND_EXECUTION]: The skill executes local shell commands and scripts to validate its setup.\n
  • Evidence: SKILL.md Step 7 directs the agent to run a provided validation script (bun scripts/check-runbooks.ts), execute newly generated scripts with the --help flag, and run existing project toolchains (typecheck, lint, tests).\n- [DYNAMIC_EXECUTION]: The skill generates new TypeScript scripts using a predefined template and then executes them to verify functionality.\n
  • Evidence: SKILL.md Step 6 and Step 7 describe the process of creating scripts from assets/script-template.ts and running them with bun.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 10, 2026, 07:45 PM