Free lead artifact · v3.4.0
The Evaluation Master Library
One download that turns your own coding agent into an auditor. It reads your repository before it recommends anything, screens thousands of catalogued capabilities and discoveries against what it found, and hands you a report you can take to code review.
No email, no account. Historical internal names inside the package are part of its history and are deliberately left untouched.
Three steps
1
Attach
Drop the unzipped kit into the repository you want evaluated and attach the catalog, the standing order, the operating rules and the report template to your coding agent.
2
Say “follow the kit”
The standing order is the prompt. The agent reads your host repository first, then screens the inventory against what it found.
3
Collect the report
A self-contained HTML report lands at your repo root, with a Markdown twin for code review. No network, prints to PDF, reads on a phone.
One run, end to end
- 1
Download
Take the free kit. No account, nothing installed.
Free
- 2
Point
Hand the kit and your repository to the coding agent you already use.
Your agent
- 3
Reject
Claims are demoted first: every finding is named and located before a fix is offered.
Verdict before repair
- 4
Report
You get a written verdict you can read, argue with, and hand to someone else.
Always
- 5
Keep
What survived becomes named, cited capability — retained only on a licensed run.
Practitioner or owned
- 01
Verdict
The evaluation is on paper before anything is written.
- 02
Your approval
Per proposal — approved, declined or deferred. Silence is never approval.
- 03
Build intent
Signed, single-use and expiring. Without it the foundry refuses.
- 04
Verification
The composed result is proven in the architecture it entered.
- 05
The gate
Cleared work arrives as a branch and a pull request you read first.
Nothing is written before the verdict, and nothing leaves the ring before the gate. A run that ends with “this already works” has still completed the circle.
Evaluate your own software in ten minutes
No account, no signup, nothing installed. You need the free ZIP, a repository, and whichever coding agent you already use — Claude Code, Cursor, Copilot, Codex, Windsurf, Cline, or a chat window you paste into.
- 0–1 min
Download and unzip
Take the 0.9 MB ZIP and unzip it into your repository as ./cmpsbl-agent-kit. Nothing is installed and nothing phones home.
- 1–2 min
Point your agent at the kit
Attach AGENTS.md, PROMPT.md and the catalog. In a chat-only setup, paste the standing order and let the agent request files as it needs them.
- 2–7 min
Let it read your repository first
The agent inventories what you have — stack, seams, hot spots, invariants — before it is allowed to look at what the Collective offers. Host first is the rule the whole method rests on.
- 7–9 min
Screening and rejection
Thousands of catalogued rows get screened against your findings. Most are rejected. A handful survive, and a few of those get composed into something specific to your codebase.
- 9–10 min
Read the verdict
A self-contained HTML report lands at your repo root: verdict, ranked recommendations, priced work, stated ceilings, and the reasons for every rejection. Then you decide whether $499 is worth it.
The whole prompt
Read AGENTS.md and PROMPT.md in ./cmpsbl-agent-kit, then follow the standing order exactly: inventory this repository first, screen the catalog against what you actually found, reject anything that does not fit, and write the integration report to the repo root. Change no source files.
Want to see the shape of the output before you run it? Read six sealed reports from real repositories — including two that argued against buying.
What a good report contains
- No surviving {{TOKEN}} placeholders — a remaining one is a defect.
- A Step 0 licence verdict for your repository — integrate, or STOP with the reason named — reached before a single call is wired.
- The commit SHA it evaluated, in the masthead — or a plain statement that it could not pin one.
- Coverage stated with numbers: rows parsed, rows matched, rows read in full, rows recommended. Sampling admitted rather than implied.
- REPLACE / HARDEN / SKIP per candidate, mapped to your own file paths.
- A SKIP list. An evaluation with no skips did not evaluate anything.
- Net LOC delta, plus each component's verification tokens quoted verbatim — every unproven axis reading “Not claimed” in those two words.
- Adapter ports named as work you must schedule — never as included.
- Net-new components designed by composing catalog pieces against your codebase.
What's in the download
PROMPT.md
The standing order: the exact instruction the agent executes with no human prompting. Attach it and say nothing more than “follow the kit”.
For: your coding agent
AGENT-HANDOFF-CATALOG.md
The merged catalog: instructions, licence rules, terms, method, the complete component catalogue, the discovery-library index, and the copy-paste prompt in PART 3.
For: your coding agent
CATALOG-PHASES.md
Router for the three-phase read path — which phase file to read when, for agents that cannot hold the merged catalog.
For: your coding agent
CATALOG-PHASE-1-DIRECTIVES.md
Phase 1 — standing order, assignment, price, licence rules, Step 0. No components.
For: your coding agent
CATALOG-PHASE-2-SHORTLIST.md
Phase 2 — class and capability indexes plus one compact row per component: id, name, mode, LOC, verification, constraints.
For: your coding agent
CATALOG-PHASE-3-COMPONENTS.md
Phase 3 — the full component rows verbatim, including what each does and its API surface.
For: your coding agent
DISCOVERY-VAULT-INDEX.md
Legacy index, kept in the tree for continuity only. The pool it describes is retired: nothing on the server reads it, its total is never quoted as library size, and an agent should not screen from it. Screen the component catalogue instead.
For: nobody — retained for continuity
INTEGRATION-REPORT-TEMPLATE.html
The report skeleton the agent must fill in — every {{TOKEN}} replaced, no external assets, opens from file://.
For: your coding agent
AGENTS.md
Standing operating rules: rule 0 (the standing order), ground rules, licence pointer, the Step 0 licence-conflict check with its host-licence verdict table, integration procedure.
For: your coding agent
ROOT-OF-TRUST.md
What the checksums prove and what they cannot: the root hash, why the seal is checksums only with no signature claimed, and how to anchor the tree out of band.
For: you and your reviewer
SECURITY-SCAN.md
The 31-rule repeatable sweep across both collections — 0 high, 19 medium delivery-wide — and exactly what that does and does not mean.
For: you and your reviewer
LICENSE
The CMPSBL® Perpetual Edition License 1.0, verbatim — the single licence covering every delivered file.
For: you and your legal reviewer
EXAMPLE-INTEGRATION-REPORT.html
A filled-in report, so you can see what a finished evaluation looks like before your agent writes one.
For: you
LICENSING.md
The licence in one page: what it permits, and what needs written permission.
For: you
docs/LICENSE-COMPATIBILITY.md
The host-licence verdict detail behind Step 0 — MIT, BSD, Apache-2.0, MPL, GPL and proprietary hosts, each with an answer.
For: you and your legal reviewer
docs/TAXONOMY.md
The canonical primitive taxonomy: the 40 primitives every capability unit is classified against.
For: your coding agent
README.md
Three steps, and how to check the delivered report.
For: you
What it does not contain
- No component source. The Evaluation Master Library evaluates; it does not ship the software.
- No network calls, no telemetry, no account, no SHPBL API.
- No AI service of ours. Your agent, your tooling, your repository, your call.
- No guarantee that anything is recommended — a clean “leave it alone” is a legitimate result.
Then the equation changes
The evaluation finds the work. The Complete Master Library™ does it.
The report names components by id. The Complete Master Library is where those components actually live — the whole Master Library Set, purchased once at $499, yours to keep under the supplied perpetual licence.