The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection via external data sources.\n
Ingestion points: External information is gathered using WebSearch and WebFetch (SKILL.md and SUBAGENT.md, Step 3) to identify migration steps.\n
Boundary markers: There are no explicit markers or safety instructions used to distinguish between the agent's logic and the data fetched from the internet.\n
Capability inventory: The agent has the authority to modify local source code, rewrite dependency configuration files, install new software packages, and execute test scripts (SKILL.md, Step 5).\n
Sanitization: The skill lacks a mechanism to verify the integrity or safety of the code snippets or installation commands suggested by researched web content before execution.\n- [COMMAND_EXECUTION]: The skill relies on executing system commands to perform its tasks.\n
It uses the jj (Jujutsu) version control system to analyze file changes.\n
It executes package management commands (e.g., npm install, pip install) to update project dependencies.\n
It invokes a separate test skill to execute local test runners and verify the upgrades.\n- [EXTERNAL_DOWNLOADS]: The skill performs external downloads when installing updated dependencies from public or private software registries (Step 5.2).

code-upgrade