cullendotdev/pi-skill-gate

Control which skills are injected into the initial system prompt with /skill-gate. Browse full skill content, toggle visibility on or off, and invoke skills — all from a keyboard-driven overlay. Skills that are not injected can still be executed manually in chat. State is persisted across sessions and injected into the system prompt before every agent run.

Features

• Complete skill coverage — uses pi’s ResourceLoader API with dynamic directory discovery to match /skill: autocomplete (including extension-contributed, project-memory, and migrated package skills)
• Skill browser — view the complete SKILL.md content (description, when-to-use, procedure, prompt body) for every skill with syntax-highlighted markdown rendering
• Yank body to clipboard — press y to copy the current skill’s full SKILL.md body to the clipboard for pasting elsewhere
• Open in $EDITOR — press o to open the current skill’s SKILL.md in $VISUAL / $EDITOR. The overlay pauses, the editor takes over the terminal, and the body cache is invalidated on return so your edits show up immediately
• Toggle visibility — enable or disable skills individually; disabled skills are excluded from the system prompt
• Search with highlighting — / searches by skill name; f performs full-text search across names, descriptions, and SKILL.md bodies. Matching text is highlighted in the sidebar, description, and rendered prompt body
• Bulk actions with confirm modals — enable/disable all visible skills at once (a/A, warning-bordered modal) or reset the current scope to defaults (r, error-bordered modal). Bulk respects the active search filter
• Persistent state — toggle state saved to ~/.pi/agent/config/skill-gate.json
• Usage analytics — every /skill:name invocation is counted and persisted; show a “Uses” column in the sidebar (toggle with u) and per-skill usage in the detail view
• System prompt injection — enabled skills are automatically inserted as `` before each agent start via the before_agent_start hook
• Respects native-disabled — skills with disableModelInvocation: true are marked with a disabled indicator and cannot be toggled
• Per-project configs — project-level overrides stored under a projects key in the same config file, keyed by absolute path. Toggle scope between global and project with g
• Fast hook — before_agent_start filters from in-memory caches (skills + config) — zero filesystem access per agent turn

Installation

Via npm (recommended)