[rig-sampler] Add s.percent schema helper for percentage values (0–100)

[github-actions]

Samples run

#	File	Turns	Notes
41	41-parse-coverage.ts	1	Uses s.record(s.object({lines: s.number, branches: s.number, ...})). Coverage percentages expressed as plain s.number — no range constraint.
42	42-json-repair.ts	1	Minimal agent. Clean schema, no repair needed.
43	43-snapshot-test-updater.ts	1	Uses s.enum("green","yellow","red") for rating and overallPct: s.number for coverage percentage — the number field lacks a 0–100 constraint that would catch model hallucinations.
44	44-flaky-test-analysis.ts	1	Array-output schema, clean run.
45	45-code-owner-suggestion.ts	1	Clean single-turn run.

All five samples completed in 1 turn with no repair loops.

Improvement: s.percent

Both sample 41 (lines, branches coverage fields) and sample 43 (overallPct) use s.number for percentage values. A percentage is semantically constrained to [0, 100], but s.number accepts any floating-point value — the harness will silently accept 102.5 or -5 without a validation error or repair turn.

This PR adds s.percent: a constrained NumberSchema with minimum: 0 and maximum: 100.

// Before:
output: s.object({ overallPct: s.number })

// After — range validated at runtime:
output: s.object({ overallPct: s.percent })

What changed

• skills/rig/rig.ts: Added s.percent via createConstrainedNumberSchema<NumberSchema>({ type: "number", minimum: 0, maximum: 100 }) alongside the existing s.positiveInt and s.nonNegativeInt helpers. Updated the file-header docstring to include it.
• src/rig.test.ts: Added a describe("s.percent") block with serialization, acceptance (0, 50.5, 100), and rejection (−1, 101) tests.

The change is additive and non-breaking: no existing API is modified.

Generated by Daily Rig Sampler · sonnet46 85.2 AIC · ⌖ 8.31 AIC · ⊞ 5.4K · ◷

[github-actions]

🧠 Matt Pocock Skills Reviewer has completed the skills-based review. ✅

[github-actions]

Skills-Based Review 🧠

Applied /tdd, /codebase-design, and /grill-with-docs — commenting with non-blocking improvement suggestions.

📋 Key Themes & Highlights

Key Themes

• Motivating samples not updated: Samples 41 and 43 are cited in the PR as the reason for s.percent, but still use s.number for percentage fields — the validation gap the PR aims to close remains open there.
• SKILL.md not updated: The prompt-context documentation doesn't surface s.percent in its helper table or prose, so AI agents generating Rig programs won't know to prefer it.
• Test boundary precision: Rejection tests use integers only; adding fractional near-boundary values (e.g. 100.001) would give stronger coverage confidence.

Positive Highlights

• ✅ Clean additive change — follows the established createConstrainedNumberSchema pattern exactly
• ✅ Good JSDoc comment explaining motivation and usage
• ✅ Well-structured tests: serialization + acceptance range + rejection, matching the pattern of s.positiveInt / s.nonNegativeInt
• ✅ Non-breaking: no existing API is modified

🧠 Reviewed using Matt Pocock's skills by Matt Pocock Skills Reviewer · sonnet46 33.7 AIC · ⌖ 4.38 AIC · ⊞ 6.3K
_{Comment /matt to run again}

[github-actions]

[/tdd] The rejection tests use integers only (-1, 101). Adding fractional near-boundary values would confirm the JSON Schema maximum/minimum exclusion works precisely.

💡 Suggested additions

expect(analyzeResponse(JSON.stringify(100.001), s.percent, "test", 1).ok).toBe(false);
expect(analyzeResponse(JSON.stringify(-0.001), s.percent, "test", 1).ok).toBe(false);

JSON Schema minimum/maximum are inclusive, so these verify the validator does not drift at float precision.

[github-actions]

[/codebase-design] The PR description explicitly identifies samples 41 and 43 as the motivation for s.percent, but neither sample was updated to use it. The helper provides value only when callers adopt it — leaving the motivating samples unchanged means the validation gap described in the PR body persists.

💡 What to update

In sample 41 (41-parse-coverage.md):

// Before
lines: s.number,
branches: s.number,

// After
lines: s.percent,
branches: s.percent,

In sample 43 (43-snapshot-test-updater.md): the generated sample doesn't seem to include percentage fields, but the PR description says overallPct: s.number — confirm which sample file contains it and update accordingly.

Additionally, samples 123, 143, and 153 each have overallPct: s.number and coverageByCategory: s.record(s.number) — these are candidates for s.percent and s.record(s.percent) as well.

[github-actions]

[/grill-with-docs] The header docstring and rig.ts JSDoc were updated, but SKILL.md's schema helper table (around line 62) still lists only s.number and doesn't mention s.percent, s.positiveInt, or s.nonNegativeInt. Agents using SKILL.md as prompt context won't know to prefer s.percent over s.number for percentage fields.

💡 Suggested SKILL.md addition

In the schema helpers table, add a row or note like:

| Constrained numbers | `s.positiveInt`, `s.nonNegativeInt`, `s.percent` (0–100) |

And in the prose guidance (around line 70), add: "Use s.percent instead of s.number for percentage fields to enforce the 0–100 range."