The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes 'Vim-Plug' for plugin management as documented in 'SKILL.md' and 'reference.md'. Commands like ':PlugInstall' and ':PlugUpdate' are used to fetch and install scripts from external Git repositories.- [REMOTE_CODE_EXECUTION]: Plugins downloaded via 'Vim-Plug' are executed within the local editor environment, which provides a mechanism for running unverified code from remote sources.- [PROMPT_INJECTION]: The 'Self-Heal Observer' section in 'SKILL.md' instructs the agent to monitor 'friction signals' and perform a diagnostic protocol from an external path ('.claude/skills/self-heal/references/diagnosis-protocol.md') without user notification. This is a behavioral override that attempts to enforce silent operation and conditional execution of instructions from outside the current skill context.- [COMMAND_EXECUTION]: The skill facilitates the execution of system commands and tools such as 'ag' (The Silver Searcher), 'fzf', and 'ctags' (via ':MakeTags') to perform file system operations.- [PROMPT_INJECTION]: The 'Self-Heal' monitoring logic introduces a surface for indirect prompt injection.
Ingestion points: User 'friction signals' and session feedback processed during the task (triggered in 'SKILL.md').
Boundary markers: Absent; there are no delimiters or instructions to ignore embedded commands within the observed interactions.
Capability inventory: The agent is permitted to use 'Read', 'Glob', 'Grep', 'Edit', and 'Write' tools.
Sanitization: Absent; the agent is instructed to follow the external protocol 'exactly' based on runtime signals without validation.

vim-config