The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Downloads and pipes a shell script directly to the command interpreter for environment initialization.
Evidence: curl -fsSL https://raw.githubusercontent.com/AnEntrypoint/gm/main/install.sh | sh -s -- spool in SKILL.md.
Context: The script is hosted on the author's repository and is part of the intended installation flow.
[REMOTE_CODE_EXECUTION]: Executes a PowerShell script from a remote URL via the iex command on Windows systems.
Evidence: irm https://raw.githubusercontent.com/AnEntrypoint/gm/main/install.ps1 | iex in SKILL.md.
[COMMAND_EXECUTION]: Executes locally installed daemon binaries from the user's profile directory.
Evidence: & "$env:USERPROFILE\.gm-tools\agentplug-runner" spool in SKILL.md.
[EXTERNAL_DOWNLOADS]: Fetches setup and configuration files from the vendor's GitHub repository.
Evidence: Interaction with raw.githubusercontent.com/AnEntrypoint/gm/main/.
[INDIRECT_PROMPT_INJECTION]: Ingests potentially untrusted data from web browsing and code search tool outputs via a local file spooler.
Ingestion points: JSON files in .gm/exec-spool/out/ populated by tool outputs (e.g., browser, fetch).
Boundary markers: Data is structured as JSON; however, the agent's instructions prioritize these outputs as ground truth.
Capability inventory: The agent has access to Write and Bash tools, allowing filesystem and process manipulation based on ingested data.
Sanitization: No explicit sanitization of tool output is described beyond the JSON structure.
[DYNAMIC_EXECUTION]: Supports the addition of custom execution hooks which grant third-party code the same authority as the project.
Evidence: Section 4 describes the use of hooks/*.js for project reconfiguration.
Context: The skill includes a mitigation requiring the agent to ask the user before enabling such hooks.

gm