The Agent Skills Directory

[OBFUSCATION]: The skill instructions contain hidden U+0001 (Start of Heading) control characters embedded within shell commands, specifically between the script path and the command name. This character serves as an invisible separator that can evade security detection systems scanning for known command names.
Evidence: python "$QA_AGENT_DIR/scripts/qa_agent.py"\u0001show-knowledge (Step 3)
Evidence: python "$QA_AGENT_DIR/scripts/qa_agent.py"\u0001safe-write-json (Step 10)
Evidence: python "$QA_AGENT_DIR/scripts/qa_agent.py"\u0001check-mojibake (Step 10)
[DYNAMIC_EXECUTION]: The skill instructions direct the agent to dynamically determine the location of its core execution script by searching through hidden directories in the user's home folder (~/.claude/skills, ~/.agents/skills, ~/.codex/skills) using a find command. This allows the skill to execute code from variable and potentially untrusted locations.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process potentially untrusted data from a local repository (source code, diffs, and documentation) to generate metadata and context files.
Ingestion points: Files are read during context collection (collect-context) and test indexing (index-existing-cases).
Boundary markers: None identified. The instructions do not include delimiters or warnings for the agent to ignore embedded instructions in the analyzed data.
Capability inventory: The skill has capabilities to write files to the .qa-agent/ directory and execute shell commands via the qa_agent.py script.
Sanitization: Not present. There is no mention of validating or escaping the repository content before it is processed or used to generate outputs.
[COMMAND_EXECUTION]: The skill heavily relies on the execution of a Python CLI tool (qa_agent.py) located at a path discovered at runtime. While the skill claims to perform QA tasks, the full behavior of this script is not inspectable as its source code is not provided.

qa-context-profiler