mrodgersjs-web/rig-marketplace/meta-install-verifier

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Install a skill AND adversarially prove its done-test can go RED before certifying the install. This is a META skill — it operates on skills in the catalog, never on the content those skills produce.

Install

npx vskill@latest install mrodgersjs-web/rig-marketplace/meta-install-verifier

SKILL.md

Meta Install Verifier

Purpose

This is a META skill: it operates on skills and their done-tests, not on content. A green that won't go red is theater — so installation is only half the job. This skill installs the skill, extracts its done-test, runs it green, then executes the skill's documented planted-failure fixture and certifies the install only if the same test goes RED on the broken input. No verified red path, no certification.

Math claim (Deviatrix MathExec)

yaml
math_claim:
expression: "certified(S) <=> (T(S, good) = 0) AND (T(S, planted_bad) != 0); separation z = 0.6745 * (g - median(B)) / MAD(B) >> 0"
symbols: [S, T, good, planted_bad, g, B, z, MAD]
assumptions:
- "the done-test T is executable in the current environment, not a prose claim"
- "the planted_bad fixture is derived from the skill's own Planted failure section, not invented at install time"
- "exit codes are honored: green = 0, red != 0; stdout text alone is never the verdict"
reference_population: "exit-code distribution B of the done-test over planted-bad fixture variants (broken inputs must cluster red, far from the green run g = 0)"
estimator: "robust_madz"
expected_result: "green run exits 0; every planted-bad variant exits non-zero with the documented RED message; install verdict CERTIFIED only when both hold"
falsifier: "any planted-bad fixture that exits 0 (the test cannot go red — BLOCK with NO-RED-PATH), or any green fixture that exits non-zero (the test is broken — BLOCK with NO-GREEN-PATH)"

Workflow

  1. Install the skill into the catalog (needle install <name> or equivalent placement); record the installed path and sha256 of SKILL.md.
  2. Parse the SKILL.md: require a Done test block and a Planted failure section; missing either → BLOCK NO-RED-PATH-DOCUMENTED.
  3. Run the done-test on the good fixture; require exit 0, else BLOCK NO-GREEN-PATH.
  4. Construct the planted-bad fixture exactly as documented (e.g. undefined symbol, mutated digest, 35σ auto-pass) and run the same test; require non-zero exit and the documented RED message, else BLOCK NO-RED-PATH.
  5. Certify: write the install record {name, path, sha256, green_exit, red_exit, red_message} and seal it via meta-proof-sealer.
  6. Record the certification event via rig-memory-os.

Rig Memory OS integration

Every certification decision is recorded via memory.record_event(kind="install_certified"|"install_blocked", name=..., red_exit=...). Skills that repeatedly fail the red-path check are proposed via memory.propose_memory(...) as untrusted gates.

Done test

bash
cd $(rig-builds)/needle-haystack && python3 artifacts/meta/mathexec_substrate.py self-test && python3 -c "
import pathlib, subprocess, sys
sk = pathlib.Path('artifacts/skills/meta-install-verifier/SKILL.md').read_text()
assert '## Planted failure' in sk and '## Done test' in sk, 'NO-RED-PATH-DOCUMENTED'
good = subprocess.run(['python3','artifacts/meta/mathexec_substrate.py','self-test'],capture_output=True)
assert good.returncode == 0, 'NO-GREEN-PATH'
bad = subprocess.run(['python3','-c','''
import sys; sys.path.insert(0, \"artifacts/meta\")
from mathexec_substrate import symbolic_check
r = symbolic_check(\"x**2 + y\", [\"x\"])
assert r[\"status\"] != \"PASS\", \"PLANTED FAILURE NOT CAUGHT\"
'''],capture_output=True)
assert bad.returncode == 0, 'NO-RED-PATH: undefined-symbol fixture was not caught'
print('INSTALL VERIFIER OK: green path green, planted undefined-symbol fixture caught')"

Planted failure

Fixture: a candidate skill whose done-test passes on a deliberately broken input — concretely, symbolic_check("x**2 + y", ["x"]) where y is undefined. The verifier MUST observe the check go RED (status != "PASS"); if it reports PASS for the undefined-symbol expression, the expected RED output is AssertionError: PLANTED FAILURE NOT CAUGHT and the install verdict is BLOCK NO-RED-PATH. A verifier that certifies a skill with no demonstrable red path is itself the failure being planted against.

Examples

  • meta-install-verifier certify meta-proof-sealer → runs its done-test green, mutates a packet digest, watches the chain verify exit non-zero, prints CERTIFIED meta-proof-sealer (red path verified).
  • meta-install-verifier certify some-skill where the done-test greps for the word "PASS" in its own output → planted fixture also prints PASS → BLOCK some-skill: NO-RED-PATH.

Install

bash
needle install meta-install-verifier

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/mrodgersjs-web/rig-marketplace/meta-install-verifier/badge)](https://verified-skill.com/skills/mrodgersjs-web/rig-marketplace/meta-install-verifier)

Something wrong with this skill or its page? Tell us.