BLASTGRID

AI coding agents (Cursor, Antigravity, Claude Code, etc.) auto-install dozens of skills into your system. Over time they pile up: unused skills eat context window tokens on every request, the agent picks wrong or conflicting instructions from stale skills, and there is no way to see what is actually being used vs what is sitting there dead. With 50+ skills across multiple agents, manual cleanup is not realistic.

BLASTGRID tracks real skill usage, shows what is dead weight, and lets you clean up in bulk. One command to vault everything unused, one command to restore if needed.

Supports 12 agents: Antigravity, Claude Code, Cursor, Gemini CLI, Codex CLI, Kiro, GitHub Copilot, Windsurf, Junie (JetBrains), Roo Code, OpenCode, and the shared .agents standard. If a directory exists on your machine, BLASTGRID picks it up automatically.

Extra paths (watch.conf)

Create ~/.blastgrid/watch.conf with one absolute path per line (shell-style ~ is OK). Lines starting with # are comments. Each path may be a file or a directory; directories are scanned recursively (.git is skipped). Those files appear in the UI under agent watch and are tracked by the same daemon backends as skills. Large trees can mean many files — keep the list focused.

Example:

# My prompts project
~/Projects/prompts/rules.md

# Whole notes folder
/home/you/notes

After editing, press F5 in the TUI or run blastgrid scan. Vault restore for single vaulted watch files uses a .blastgrid-origin marker inside the vault folder.

Install

python3 -m venv .venv && source .venv/bin/activate && pip install -e .

After that: blastgrid to launch, or ./run.sh (activates venv for you).

How it works

1. On launch, BLASTGRID scans all agent skill directories for SKILL.md files, estimates token cost of each skill, and stores everything in a local SQLite database.

2. A background daemon tracks which skills are actually read by agents. It detects file access using filesystem timestamps (atime). When an agent reads a SKILL.md, the daemon logs it and increments the usage counter.

3. After running for a while, you see which skills are actively used and which are dead weight (ghosts). Press A to open AUTOPWN — enter a threshold (e.g. 5), and every skill with fewer uses gets moved to a backup vault. Protected skills (tagged SECURED or KEEP) are excluded.

4. Vaulted skills can be restored at any time with Z or blastgrid restore.

Tracking details

The core trick: APFS (macOS) and ext4 (Linux) with relatime only update atime when atime < mtime. The daemon sets atime = mtime - 1 day on each SKILL.md. When an agent reads the file, the kernel bumps atime — the daemon catches the change and logs it, then re-primes for the next read.

macOS:

• Without sudo — atime-poll backend. Polls every 1s using the priming trick above. Works on APFS out of the box.
• With sudo — fs_usage backend. Hooks into the kernel's file activity stream. Most precise, catches every read in real-time.

Linux (Ubuntu, etc.):

• With inotify-tools — inotifywait backend. Kernel-level access event monitoring, real-time. Install: sudo apt install inotify-tools.
• Without inotify-tools — falls back to atime-poll, same priming trick as macOS. Works on ext4 with relatime (Ubuntu default).
• If the disk is mounted with noatime, the daemon will detect this and suggest either installing inotify-tools or remounting with relatime.

Quick start

blastgrid              # TUI with live dashboard
blastgrid daemon       # standalone monitoring with timeline
blastgrid stats        # quick summary
blastgrid top 100      # top 100 skills by usage
blastgrid agents       # detected agents and paths
blastgrid restore      # list / restore backed-up skills
blastgrid help         # all commands

TUI

Four tabs, switch with 1 2 3 4:

• Dashboard — context budget bar, live activity chart (10 min), session usage tiers, agent breakdown
• Armory — full skill inventory with search, sorting by uses/tokens/agent, tagging
• Hunt — ghost tracker: unused skills on top (sorted by token weight), active below. Cleanup hub
• Secured — skills immune to AUTOPWN and mass actions

Key bindings:

Key	Action
A	AUTOPWN — mass vault by usage threshold
S	Toggle SECURED (immune to mass actions)
Z	Restore all from vault
B	Vault selected skill
K / R / C	Tag keep / remove / clear
X / U / P	Mark / unmark / purge ghosts
ENTER	Skill detail modal
TAB	Cycle agent filter
F5	Re-scan

Data

~/.blastgrid/:

File	Purpose
blastgrid.db	Skills, usage counts, tags (SQLite)
watch.conf	Optional: extra files/dirs to track (one path per line)
vault/	Backed-up skills (restorable)
graveyard/	Deleted skills
usage.log	JSON lines access log (10 MB rotation)

Supported agents

Agent	Global skill path
Antigravity	~/.gemini/antigravity/skills/
Claude Code	~/.claude/skills/
Cursor	~/.cursor/skills/, ~/.cursor/skills-cursor/
Gemini CLI	~/.gemini/skills/
Codex CLI	~/.codex/skills/
Kiro	~/.kiro/skills/
GitHub Copilot	~/.copilot/skills/
Windsurf	~/.codeium/windsurf/skills/
Junie (JetBrains)	~/.junie/skills/
Roo Code	~/.roo/skills/
OpenCode	~/.config/opencode/skills/
Shared	~/.agents/skills/
watch.conf extras	Paths listed in ~/.blastgrid/watch.conf (any files; agent column shows watch)

Run blastgrid agents to see what's detected on your machine.

Requirements

Python 3.10+. Dependencies (textual, rich) install automatically.