{
  "markdown": "# Tokentamer\n\nAudits a project's Claude Code session transcripts to find concrete, evidence-backed opportunities to have used fewer tokens: repeated/duplicated work, context pollution, unused MCP tools, poorly-disclosed skills, bloated prompts, verbose CLAUDE.md/memory files, wrong model choices, missed memory-save opportunities, and places a deterministic script would have beaten an LLM call.\n\n## Highlights\n\n- **Evidence-based findings** — every finding cites a session id, timestamp, and quote or tool-call sequence, not generic advice\n- **Deterministic extraction** — a bundled script parses transcript JSONL directly instead of asking the model to eyeball raw logs\n- **Ten review categories** — context pollution, duplicated work, LLM-vs-script calls, idle MCP tools, undisclosed skills, oversized prompts, verbose harness files, wrong model choice, missed memory opportunities, and redundant reads\n- **Scales to large histories** — output is newline-delimited JSON (one session per line), so a large project can be paged through or delegated to subagents instead of loaded whole into context\n- **Cross-platform** — ships as a native skill for Claude Code, Codex, Antigravity, and Gemini CLI, with documented fallbacks where a platform lacks a feature the others have\n\n## Getting Started\n\n### Prerequisites\n\n- Node.js (v18+)\n\n### Installation\n\n#### From the mackayi marketplace (recommended)\n\n```\n/plugin marketplace add keithmackay/mackayi\n/plugin install tokentamer@mackayi\n```\n\n#### Claude Code\n\n```bash\ncp -r /path/to/tokentamer/ ~/.claude/skills/tokentamer/\n```\n\nOr symlink:\n```bash\nln -s /path/to/tokentamer/ ~/.claude/skills/tokentamer\n```\n\nThen invoke with: `/tokentamer`\n\n#### Codex\n\nPlace the plugin directory where Codex can find it, then add an entry to your marketplace:\n\n**`~/.agents/plugins/marketplace.json`** (create if absent):\n```json\n{\n  \"name\": \"personal\",\n  \"interface\": { \"displayName\": \"Personal Plugins\" },\n  \"plugins\": [\n    {\n      \"name\": \"tokentamer\",\n      \"source\": { \"source\": \"local\", \"path\": \"/path/to/tokentamer/\" },\n      \"policy\": { \"installation\": \"AVAILABLE\", \"authentication\": \"ON_INSTALL\" },\n      \"category\": \"Productivity\"\n    }\n  ]\n}\n```\n\n#### Antigravity\n\n**Global install** (all workspaces):\n```bash\ncp -r /path/to/tokentamer/ ~/.gemini/antigravity/skills/tokentamer/\n```\n\n**Workspace install** (current project only):\n```bash\ncp -r /path/to/tokentamer/ .agents/skills/tokentamer/\n```\n\nThe root `SKILL.md` has no Claude Code-specific frontmatter (no `metadata`, `retrieval`, or `tags` fields), so it is used as-is — no separate `antigravity/SKILL.md` was needed.\n\nSkills are auto-discovered. You can also mention the skill by name to force activation.\n\n#### Gemini CLI\n\nGemini CLI installs extensions directly from GitHub:\n\n```bash\ngemini extensions install https://github.com/<owner>/tokentamer\n```\n\nTo update:\n```bash\ngemini extensions update tokentamer\n```\n\nThe skill is auto-discovered from `GEMINI.md` after installation. Local install is not directly supported — this directory must live in a GitHub repository for `gemini extensions install` to work.\n\n## Usage\n\nInvoke the skill directly (`/tokentamer` on Claude Code, or the equivalent trigger on other platforms) and give it a project to audit:\n\n```\n/tokentamer review ~/Projects/my-app for token waste\n```\n\nThe skill locates that project's transcripts under `~/.claude/projects/`, runs the bundled scanner, checks each finding against the category list in `references/categories.md`, and writes a report using the structure in `references/report-template.md`.\n\nYou can also run the scanner directly to inspect the raw data yourself:\n\n```bash\nnode scripts/scan-transcripts.js /Users/you/Projects/my-app > /tmp/scan.ndjson\n```\n\n```bash\n# Untruncated prompt text, one session only\nnode scripts/scan-transcripts.js /Users/you/Projects/my-app --full --session <sessionId>\n```\n\nEach line of output is one session's worth of user turns (truncated to 400 chars by default), per-session model usage, tool-call counts, MCP tool-call timestamps, and Skill invocations — the same data the skill itself analyzes.\n\nRun `/tokentamer --help` (or the equivalent trigger on other platforms) to print usage without running the workflow.\n\n## Development\n\nThis is a documentation-and-script skill package, not a compiled project — there's no build step or test suite to run.\n\n```bash\ngit clone <repo>\ncd tokentamer\nnode scripts/scan-transcripts.js \"$PWD\" --session <sessionId>   # sanity-check the scanner against this repo's own transcripts\n```\n\nWhen editing `SKILL.md`, keep it under ~500 words (progressive disclosure — move heavy reference material into `references/`) and re-sync the ported copies under `skills/tokentamer/` for Codex and Gemini CLI so all four platform versions stay consistent. Run `scripts/check-sync.sh` to verify the shared `references/*.md` and `scripts/scan-transcripts.js` files still match between the two trees (it deliberately skips `SKILL.md` and `help.md`, which differ by design between platforms).\n\n`docs/reviews/` and `docs/plans/` are internal dev-history artifacts from past `/improve-this` review passes on this repo, kept for context — they aren't user-facing documentation.\n\n## Contributing\n\nContributions are welcome — fork the repo, make your changes on a branch, and open a pull request. If you're changing `SKILL.md` or `references/`, please also update the corresponding copy under `skills/tokentamer/` so the Codex and Gemini CLI ports don't drift out of sync.\n\n## Changelog\n\nSee [CHANGELOG.md](CHANGELOG.md) for release history.\n\n## License\n\n[MIT](LICENSE)\n\n## Compatibility\n\n| Feature | Claude Code | Codex | Antigravity | Gemini CLI |\n|---------|:-----------:|:-----:|:-----------:|:----------:|\n| Core skill | ✅ | ✅ | ✅ | ✅ |\n| Sub-documents (`references/`) | ✅ | ✅ | ✅ | ✅ |\n| Scripts (`scripts/scan-transcripts.js`) | ✅ | ✅ | ✅ | ✅ |\n| `superpowers:writing-skills` reference (plugin namespacing) | ✅ (optional; graceful fallback if absent) | ❌ | ❌ (not installed) | ❌ |\n| `artifact-design` skill reference (report publishing) | ✅ (optional; graceful fallback if absent) | ❌ | ❌ (not installed) | ❌ |\n| Subagent dispatch (per-session delegation) | ✅ | ✅ | ✅ | ❌ |\n\nLegend: ✅ Supported · ❌ Not supported\n\nWhere a Claude Code-specific skill reference has no equivalent, the ported `SKILL.md` documents a fallback under **Platform Limitations** (Codex and Gemini CLI copies) rather than failing silently. On Claude Code itself, both `superpowers:writing-skills` and `artifact-design` are optional — the workflow degrades gracefully if either plugin isn't installed.\n\n## References\n\n- **Claude Code Skills:** https://code.claude.com/docs/en/skills\n- **Claude Code Complete Guide (PDF):** https://resources.anthropic.com/hubfs/The-Complete-Guide-to-Building-Skill-for-Claude.pdf\n- **Codex Plugins:** https://developers.openai.com/codex/plugins/build\n- **Antigravity Skills:** https://antigravity.google/docs/skills\n- **Gemini CLI Extensions:** https://github.com/google-gemini/gemini-cli/blob/main/docs/extension.md\n- **Agent Skills open standard:** https://agentskills.io/home\n",
  "bytes": 7120,
  "sha": "df4da4f532d296d320fc5ea5113f15c453b6b7982a75fdb76eed454a652e6b68",
  "repo_slug": "keithmackay/tokentamer",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_keithmackay_tokentamer_08521104/readme"
}