100xprism

---

Quick start

1. Install (once per machine).

# npm — macOS, Linux, Windows (needs Node.js 18+)
npm install -g 100xprism && 100xprism install

# or curl — macOS / Linux
curl -fsSL https://raw.githubusercontent.com/rajitsaha/100xprism/main/get.sh | bash

Both clone the toolkit to ~/100xprism, set up Claude Code, and let you toggle components (modules, plugins, templates, enforcing hooks). The npm package is a thin launcher; modules, hooks and plugins live in that clone.

2. Set up a project (once per repo).

cd your-project && 100xprism init

init asks which tools you use in the repo (Claude Code, Cursor, Codex, Pi) and writes their files. Commit them so teammates get the same setup on clone.

3. Use it. Restart Claude Code, then run:

/gate      # run the quality gate
/commit    # gate → stage → conventional commit

Two places, two kinds of command. 100xprism … runs in your terminal. /gate, /commit and the other slash commands run inside your AI tool.

---

What you get

---

Lean by default: what sits in your context

A module's description is re-sent to the model on every turn; its body loads only when used. So every installed module costs tokens all the time. 100xPrism installs the minimum by default:

The 12 must-have workflows: gate commit push branch pr release test lint security docs eval pair-loop.

Reaching a module that isn't installed. Use the generic route, or just describe the task — the always-loaded 100x-resolver tells the agent where each catalog module lives:

/100x spec               # run the spec workflow
/100x fix-bugs "login returns 403"

Changing the mode. optimize writes your choice and re-emits immediately. Run it from a project root — it also updates that project's generated files.

100xprism optimize --skills=profile   # every slash command becomes native (/spec, /fix, …)
100xprism optimize --skills=all       # every module installed
100xprism optimize                    # back to must
100xprism optimize --dry-run          # preview without changing anything

Estimated always-on description cost (characters ÷ 4, checked in CI — not billed usage):

Details: docs/token-optimization.md.

---

Slash commands

All 28, grouped by purpose. ★ = installed natively in the default must mode. Run the others as /100x <slug> (slug in brackets where it differs), or switch to --skills=profile to get them all natively.

Lifecycle

Quality

Engineering

Data, docs & architecture

Task-triggered playbooks (40). No command — describe the task ("audit SEO on this page") and the agent loads the matching playbook: from the catalog via 100x-resolver by default, or as its own skill in all mode.

In Codex use skills by name ($gate, $commit) or pick from /skills. In Pi use /skill:gate. In Cursor, ask for the workflow by name ("run the gate workflow").

---

How each tool receives the modules

install and update run the Hermes adapter automatically when ~/.hermes or the hermes CLI exists. If you install Hermes later, run 100xprism hermes (or wait for your next 100xprism update).

Claude Code plugins are Claude-only. Use Codex /plugins for Codex-native plugins.

Windows: install, init and update work natively for Claude Code, Cursor and Codex. Pi, Hermes, and background token collection need macOS or Linux (WSL works).

---

Keeping up to date

100xprism check                   # is a newer version available?
100xprism update                  # pull latest; add, update and remove generated files
npm install -g 100xprism@latest   # (optional) update the launcher itself

update reconciles: it adds new modules and plugins, updates changed ones, and removes ones that were deleted or merged upstream — in ~/.claude and in every project you ran init in (listed in ~/.100xprism/tracked-projects). Restart Claude Code after updating.

What 100xPrism will and won't delete. It removes only files it can prove it wrote: a generated marker, a manifest entry, a /100xprism/hooks/ path, or an exact v1 file header. Your own skills, commands, hooks, plugins and a hand-written AGENTS.md are never touched. Files removed from a project are first copied to ~/.100xprism/removed-artifacts/<timestamp>/.

Repairing and removing.

100xprism audit                  # report standing context and drift
100xprism audit --fix            # repair dead hook paths, v1 leftovers, missing catalog
100xprism uninstall              # remove skills, aliases, hooks, catalog, links (keeps your data)
100xprism uninstall --projects   # …also generated files in every tracked project
100xprism uninstall --purge      # …also all of ~/.100xprism, including token history

Run 100xprism uninstall before npm rm -g 100xprism. npm's own uninstall hook only stops the dashboard and removes hooks that point into the deleted package.

---

Optional vendor skill packs

Packs are skill sets that matter only to some projects, so they never install by default. Manage them from inside Claude Code:

/pack                    # list packs, install state, and what matches this project
/pack detect             # only packs that match this project
/pack add databricks     # install
/pack remove databricks  # remove only what 100xPrism installed

Where the vendor ships an installer (e.g. databricks aitools install), the pack uses it. Available today: Databricks (Unity Catalog, Asset Bundles, Lakeflow, model serving, vector search).

---

Token, delivery & value economics

Agents spend real money. The dashboard keeps three questions apart: what did it cost, what was delivered, and what was it worth — the last stays "Not measured" until you connect an outcome source. It never invents a value score.

100xprism tokens            # open the local dashboard (http://127.0.0.1:8787)
100xprism tokens --print    # text summary, no server
100xprism tokens --json     # machine-readable report (filter with --tool codex)
100xprism value             # git delivery report for the current directory

• Cost. Claude Code and Codex record exact token counts locally; cost is priced per model from a dated catalog (scripts/pricing.py). Pi is counted when its records carry token counts. Cursor and Antigravity expose no token counts locally, so they are shown as activity only, with cost — (never a made-up $0).
• Delivery. Commits, merged PRs, releases and churn from git, joined to the spend that could be attributed to each directory.
• macOS: install adds a LaunchAgent that collects every minute into a private SQLite store (~/.100xprism/token-economics/), so history survives deleted transcripts. The dashboard needs a per-launch token (the command opens it for you) and stops after 10 idle minutes. Linux/Windows: reports are computed on demand.
• Nothing starts from your shell. Nothing leaves your machine unless you opt in to GitHub PR insights.

Budgets, GitHub insights, content search and every flag: docs/USAGE.md › Monitoring token usage.

---

Built with 100xPrism

What it changed

Snapshot taken 2026-08-02 from each repo's git history.

Across all five products, open and closed source: 3,282 commits · 1,225 merged PRs · 1.78M lines added, by one person.

As an estimate, not a measurement. Nothing was built twice, so there is no counterfactual. Assume a conventional team merges ~2 feature-sized PRs per engineer-week: 1,225 ÷ 2 ≈ 612 engineer-weeks ≈ 11.8 engineer-years. At $85/hr that is roughly $2.1M. PR size varies widely, and assuming 1 or 4 PRs per week moves the figure by 2×. Revenue, retention and customer outcomes are not measured here.

---

Common CI traps it fixes

npm install 404 inside Docker · useState(false) opacity-0 breaking Playwright · integration tests silently skipped. Full breakdown →

---

More

• Usage guide — per-tool details, project config, CI templates, dashboard, troubleshooting
• Token optimization — retention classes, modes, measuring footprint
• Enforcing hooks — gate-on-commit, secret scan, and their switches
• Architecture — why modules replaced workflows + skills
• Changelog · Roadmap · Security · Issues
• Contributing: start with AGENTS.md

---