Back to search
skilltally
MarketplaceWorks with Claude CodeGitHub
Real-time skill usage tracking via hooks for Claude Code, Codex, Antigravity & Gemini CLI.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Checked today
Badge for your README
Markdown
[](/listings/112516101106--skilltally)HTML
<a href="/listings/112516101106--skilltally"><img src="/api/listings/112516101106--skilltally/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
# 1. Detect installed agents and register lightweight PostToolUse/AfterTool hooks npx skilltally install --dry-run npx skilltally install # 2. Import historical sessions from local logs npx skilltally backfill # 3. Verify hook registration and run live self-test npx skilltally doctor # 4. Inspect usage, insights, HTML dashboard, and unused context tax npx skilltally stats --since 30d npx skilltally insights --since 30d npx skilltally dashboard --open npx skilltally unused --since 90d
Or ask your coding agent directly: *"Which agent skills do I use the most?"*
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 18 lines
- {
- "$schema": "https://anthropic.com/claude-code/marketplace.schema.json",
- "name": "skilltally",
- "description": "Real-time skill usage tracking via hooks for Claude Code, Codex, Antigravity & Gemini CLI.",
- "owner": {
- "name": "112516101106",
- "url": "https://github.com/112516101106"
- },
- "plugins": [
- {
- "name": "skilltally",
- "description": "Track skill activations and identify unused skills eating context.",
- "version": "1.2.1",
- "source": "./",
- "category": "productivity"
- }
- ]
- }
README.md · README · 158 lines
- # skilltally
- [](https://github.com/112516101106/skilltally/actions/workflows/ci.yml)
- [](LICENSE)
- [](package.json)
- [](package.json)
- Real-time skill usage tracking via hooks for **Antigravity**, **Claude Code**, **Codex**, and **Gemini CLI** — 100% local, zero dependencies.
- Find out which agent skills you actually use, and how much context (`KB` / `~tokens/session`) unused skills and plugins cost you on every turn.
- ## Quick Start
- ```bash
- # 1. Detect installed agents and register lightweight PostToolUse/AfterTool hooks
- npx skilltally install --dry-run
- npx skilltally install
- # 2. Import historical sessions from local logs
- npx skilltally backfill
- # 3. Verify hook registration and run live self-test
- npx skilltally doctor
- # 4. Inspect usage, insights, HTML dashboard, and unused context tax
- npx skilltally stats --since 30d
- npx skilltally insights --since 30d
- npx skilltally dashboard --open
- npx skilltally unused --since 90d
- ```
- Or ask your coding agent directly: *"Which agent skills do I use the most?"*
- ## Supported Agents
- | Agent | Real-time hook | History backfill | Plugin / Extension install | Notes |
- |---|---|---|---|---|
- | **Antigravity** | `PostToolUse` (`read_file\|view_file\|run_command`) | `~/.gemini/antigravity*/brain/` | Native plugin (`plugin.json` + `hooks.json`) | POSIX fast-path forwards `SKILL.md` & `skills/*/(scripts\|bin)/*` (~7 ms skip on misses); preserves `<SKILL>` manual triggers (`via: "slash_command"`); reads `transcript_full.jsonl` when truncated |
- | **Claude Code** | `PostToolUse` (`Skill\|Read\|View\|Bash`, `async: true`) | `~/.claude/projects/**/*.jsonl` | `.claude-plugin/plugin.json` | 30 s deduplication window, continuation-read filtering, and `Base directory for this skill:` slash-command backfill |
- | **Codex** | `PostToolUse` + `UserPromptSubmit` (`async: true`) | `~/.codex/{sessions,archived_sessions}/` | `.codex-plugin/plugin.json` | Supports `.jsonl.zst` natively on Node `>= 22.15`, `.system`/nested/MCP `skill://`/script detection, forked rollout deduplication, and trust verification via `/hooks` |
- | **Gemini CLI** | `AfterTool` (`activate_skill\|read_file\|view_file\|read_many_files\|run_shell_command`) | `~/.gemini/tmp/*/chats/*.{json,jsonl}` | `gemini-extension.json` + `install --host gemini` | `timeout: 5000` ms in `~/.gemini/settings.json`; supports `.json` and `$set.messages` `.jsonl` checkpoints |
- ## Plugin & Skill Installation
- If you prefer native plugin managers over `npx skilltally install`:
- - **Antigravity**: Copy or symlink this repository to `~/.gemini/config/plugins/skilltally` (or run `npx skilltally install --host antigravity`).
- - **Claude Code**:
- ```text
- /plugin marketplace add 112516101106/skilltally
- /plugin install skilltally@skilltally
- ```
- - **Codex**:
- ```bash
- codex plugin marketplace add https://github.com/112516101106/skilltally.git
- codex plugin add skilltally@skilltally
- ```
- Run `codex`, open `/hooks`, and trust the `skilltally` hooks (`PostToolUse` and `UserPromptSubmit`).
- - **Gemini CLI**:
- ```bash
- gemini extensions install https://github.com/112516101106/skilltally
- npx skilltally install --host gemini
- ```
- - **Companion skill only (`skills.sh`)**:
- ```bash
- npx skills add 112516101106/skilltally
- ```
- ## Example Output
- ### `skilltally stats --since 30d`
- ```text
- skilltally — last 30d (2026-10-20 -> 2026-11-19)
- hosts: antigravity 35 · codex 12
- triggers: auto 38 · manual 9
- # skill activations reads last used hosts trend (14d)
- 1 requesting-code-review 8 10 2026-11-12 14:03 antigravity, codex ··▁·▃·█··▃·▁··
- 2 systematic-debugging 5 9 2026-11-10 09:41 codex ····▁··█··▁···
- 3 agent-browser 3 4 2026-11-05 12:18 antigravity, codex ··█···········
- installed: 214 · used: 22 · unused: 192 -> run `skilltally unused`
- ```
- ### `skilltally unused --since 90d`
- ```text
- Unused enabled skills — last 90d (2026-08-21 -> 2026-11-19): 192 (48.2 KB ≈ ~12,340 tok/session)
- context tax: [████████████░░░░] 76% unused (~12,340 / ~16,200 tok/session)
- plugin:science 47 skills 14.8 KB ~ 3818 tok/session
- alphagenome-atlas-website-links, alphagenome-single-variant-analysis, … (+41)
- personal 47 skills 10.6 KB ~ 2729 tok/session
- architecture-blueprint-generator, auditing-python-security, … (+41)
- Tip: disabling an unused plugin removes all of its skill descriptions from every session context.
- ```
- ## CLI Reference
- | Command | Description | Key flags |
- |---|---|---|
- | `install` | Install hooks for detected hosts (`runtime/track.mjs` + backups) | `--host all\|antigravity\|claude\|codex\|gemini`, `--dry-run`, `--force`, `--json` |
- | `uninstall` | Remove `skilltally` hooks (preserves all other user hooks) | `--host ...`, `--purge`, `--dry-run`, `--json` |
- | `doctor` | Check Node version, hook paths, Codex trust, events, and live self-test | `--json` |
- | `backfill` | Import historical skill reads/invocations from local logs | `--host ...`, `--since DATE`, `--dry-run`, `--json` |
- | `stats` | Usage table sorted by `activations` with 14-day sparklines and trigger ratio | `--since 30d\|all\|YYYY-MM-DD`, `--host`, `--project`, `--by-source`, `--metric reads`, `--json` |
- | `top` | Top `N` skills | `-n 20`, plus `stats` flags |
- | `unused` | Enabled installed skills with 0 activations and Context Tax breakdown | `--since 90d`, `--host`, `--all`, `--json` |
- | `timeline` | ASCII usage bar chart over time | `--skill <name>`, `--by day\|week\|month`, `--since`, `--host`, `--project`, `--json` |
- | `chains` | Skill transition chains (`A -> B`), 3-step workflows, and co-occurring pairs | `--since`, `--host`, `--project`, `-n 20`, `--json` |
- | `insights` | Runtime tokens & prompt cache hit rate, task durations (`slowestSkills`), heavy `SKILL.md` bodies (`>= 1,500` tokens), opt-in redacted prompt patterns, Context ROI, manual-only triggers, single-project scoping, and model breakdown | `--since`, `--host`, `--project`, `--json` |
- | `dashboard` | Generate self-contained offline HTML dashboard (5-metric KPI strip with `Lifetime tokens`, `Peak tokens`, `Longest task`, streaks; SVG Treemap; cache hit rate; workflow chains; Mermaid generator) | `-o <file>`, `--since`, `--host`, `--project`, `--open`, `--json` |
- | `prune` | Safely disable 100%-unused plugins or specific unused skills (with backups) | `--since 90d`, `--host all\|antigravity\|codex`, `--plugin <name>`, `--skill <name>`, `--dry-run`, `--json` |
- | `export` | Export deduplicated events as JSON or RFC-4180 CSV | `--format json\|csv`, `-o <file>`, `--since`, `--host`, `--project`, `--include-failed` |
- | `inventory` | List installed skills, sources, hosts, enabled state, and description sizes | `--no-project`, `--json` |
- | `mcp` | Start zero-dependency stdio MCP server (`skilltally_*` tools) | — |
- | `debug` | Toggle opt-in raw stdin logging (`on \| off \| status`) | `--json` |
- ## Privacy, Performance & Reliability
- - **100% local, zero telemetry**: Zero network requests, zero third-party runtime dependencies.
- - **No raw arguments or prompts stored**: Only the matched `SKILL.md` path (or `Skill`/`activate_skill` name) and `basename(cwd)` are written to `~/.skilltally/` (`0700` dir, `0600` files). Full `cwd` is omitted unless `store_cwd: true` is set, and redacted prompt previews (`80` chars, secrets/emails/paths stripped) are omitted unless `store_prompts: true` is set in `~/.skilltally/config.json`.
- - **Safe by contract**: `track.mjs` never throws, always exits `0`, and always writes `{}` to `stdout` so hooks never interrupt an agent turn.
- - **Accurate telemetry & deduplication**: Filters paginated continuation reads, deduplicates repeated same-session triggers within 30 s while preserving manual triggers (`slash_command`/`mention`), deduplicates per-turn token snapshots, merges overlapping turn duration intervals, and computes provider-aware prompt cache hit rates across Anthropic vs. OpenAI/Gemini token schemas.
- - **Automatic rotation**: When `events.jsonl` exceeds 20 MB, it rotates automatically to `events-YYYY-MM.jsonl` while reports transparently read all archives.
- - **Config backups**: `install` and `uninstall` back up modified host configs to `~/.skilltally/backups/`.
- ### Configuration (`~/.skilltally/config.json`)
- ```json
- {
- "exclude_skills": ["skilltally"],
- "exclude_sources": [],
- "aliases": { "native_swarm": "swarm" },
- "store_cwd": false,
- "store_prompts": false
- }
- ```
- ## Troubleshooting
- - **No events recorded?** Run `npx skilltally doctor` to verify hook paths and run the live pipe self-test.
- - **Codex hooks not firing?** Run `codex` and open `/hooks` to review and trust the `skilltally` hooks (`PostToolUse` and `UserPromptSubmit`).
- - **Gemini CLI hooks disabled?** Check `"hooksConfig": { "enabled": true }` in `~/.gemini/settings.json`.
- - **Inspecting unknown hook payloads:** Temporarily run `npx skilltally debug on`, trigger a skill once, inspect `~/.skilltally/debug.jsonl`, then run `npx skilltally debug off` (which deletes `debug.jsonl` automatically).
- ## Development
- ```bash
- npm test
- npm run check:pack
- ```
- ## License
- MIT — [112516101106/skilltally](https://github.com/112516101106/skilltally)
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.