The Agent Skills Directory

[PERSISTENCE]: The skill includes an installation script scripts/install-dm.sh designed to modify the user's shell configuration files (~/.zshrc, ~/.bashrc, or ~/.profile). It appends a code block that sources scripts/dm.sh, establishing a persistent shell alias (dm) that remains active across different terminal sessions.
[EXTERNAL_DOWNLOADS]: The instructions in SKILL.md guide the user to install external dependencies, including the portless NPM package (npm install -g portless) and an additional agent skill melech-live-browser from the author's GitHub repository (https://github.com/AdirD/agent-shell-hamelech).
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow where the agent reads logs (events.jsonl) collected from a local UI via the dm logs command.
Ingestion points: The logs_session function in scripts/debug_session.py reads and parses JSONL data from the events.jsonl file.
Boundary markers: The instructions do not specify any boundary markers or delimiters when presenting the log data to the agent, creating a surface for embedded instructions in the UI to influence the agent.
Capability inventory: The skill has the ability to manage background processes using subprocess, delete directories, and is expected to modify application code based on the evidence gathered.
Sanitization: While assets/collector.py contains a SENSITIVE_KEYS filter to block credentials (e.g., api_key, token, password), it does not sanitize log content against malicious natural language instructions.
[COMMAND_EXECUTION]: The management script scripts/debug_session.py uses subprocess.Popen and subprocess.run to launch the portless proxy and the Python collector script, and to manage process lifecycles using system tools like ps and taskkill. The arguments for these calls are constructed using randomized tokens and validated session paths.

melech-debug-mode