AB miroprism
MiroPRISM — Adversarial two-round review protocol. Extends PRISM with a mandatory second round where every reviewer must respond to all R1 findings with evidence requirements enforced by a structured anti-herding guardrail. Reduces cascade sycophancy: reviewers cannot agree with a finding without independent evidence, cannot change their verdict without citing cause, and can mark findings UNCERTAIN rather than force a weak call. Core insight: A finding that survives explicit challenge is more reliable than one that was never challenged.
As a process B 71/100 · Nearly there — weak spots: inputs and preconditions, execution cost
How to improve
- The SKILL.md body is over 5,000 tokens: move reference detail into references/ and load it when needed.
- Your own cases (evals/evals.json, 4–6 real requests with expected answers): the full check would then run those instead of a model-drafted suite.
- A spec.yaml with trigger phrases and assertions — a behaviour contract for CI; `skilltest init` writes a template.
Guard findings · 0
✓ No critical or high findings
Files scanned: 5. Evidence is masked. Grey chips explain why severity was lowered.
Against the Agent Skills spec
- warning
body-longSKILL.md body ≈ 9031 tokens (recommended < 5000); move details to references/
Process rating: all ten parameters 71/100
- 0Inputs and preconditions. Does not say what the process needs to start
- 40Execution cost. Instruction body is 9031 tokens: crowds the task out of the window
- 50When it triggers. No condition that starts the skill
- 60Result and completion. Output format stated, no completion criterion
- 70Failures and branches. 9 branches
- 100Tools and files. No external tools needed
- 100Steps. 86 steps
- 100Consistency. Name and required fields are in place
- 100Running it twice. Mutating operations check current state
- 100Progress reporting. Reports progress
- medium Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- low 16 top-level sections: this looks like several domains in one skill
- low The response is described with custom markup (4 tags): a typed call is more reliable
Everything here is measured from the skill text rather than judged by a model, so the numbers are checkable. A parameter weighs more when it is a more common reason for the process to stall.
Quality signals
- +5Description has no quoted example phrases that should trigger the skill
- +4Description does not say when NOT to use the skill (false activations)
- -226 emoji in the instructions: noise for the model
- +2Single-language instructions
- +3Description length 542: enough signal without eating the budget
- +4Structure: 26 headings
- +3Step-by-step instructions: 86 items
- +3Output format is stated explicitly
- +4Has examples (27 code blocks)
- +4Reference files are cited in the instructions (1 of 1)
- +1License stated
Quality base 70; lint remarks subtract, signals add up to 100. Result: 80.