[rig-tasks] Add 10 rig samples — 2026-07-28

[github-actions]

Summary

Added 10 new rig sample files to skills/rig/samples/.

#	File	Description	Typecheck
1	270-dotfile-inventory-mapper.md	Dotfile inventory mapper using p.bash + categorizeDotfile tool	pass
2	271-regex-pattern-tester.md	Regex pattern test runner with input schema and runRegexTest tool	pass
3	272-commit-churn-classifier.md	Commit churn classifier using p.bash git log + steering addon	pass
4	273-npm-package-size-estimator.md	NPM package size estimator with classifySize tool	pass
5	274-ts-branch-coverage-analyzer.md	TypeScript branch coverage analyzer with subagent delegation	pass
6	275-env-variable-type-inferrer.md	Env variable type inferrer using p.readOptional + inferType tool	pass
7	276-git-hook-installer.md	Git hook installer with async defineTool + repair addon	pass
8	277-ts-type-guard-generator.md	TypeScript type guard generator with async extractInterfaces tool	pass
9	278-json-fixture-anonymizer.md	JSON fixture anonymizer with s.unknown param in defineTool	pass
10	279-csv-to-markdown-table.md	CSV to Markdown table converter with parseCSVRow tool + repair	pass

Typecheck failures

No failures — all 10 tasks passed typecheck.

Tasks run

• (reused) Dotfile inventory mapper using p.bash find, repair addon, s.record output with dotfiles + categorySummary
• (reused) Regex pattern test runner with input s.object, nested arrays, allPassed boolean
• (reused) Commit churn classifier with steering addon, s.record(s.object) root output
• (reused) NPM package size estimator with p.bash npm pack --dry-run, classifySize tool
• (reused) TypeScript branch coverage analyzer with subagent + p.readInput
• (reused) Env variable type inferrer with p.readOptional, regex-based inferType tool
• (new) Git hook installer with async defineTool, .git/hooks check, repair addon
• (new) TypeScript type guard generator with async extractInterfaces, p.readInput, p.write
• (new) JSON fixture anonymizer with s.unknown defineTool parameter, p.readInput + p.write
• (new) CSV to Markdown table converter with parseCSVRow tool, s.boolean input, repair addon

Generated by Daily Rig Task Generator · sonnet46 104.3 AIC · ⌖ 9.35 AIC · ⊞ 6.7K · ◷

[github-actions]

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

[github-actions]

Skills-Based Review 🧠

Applied /codebase-design — no blocking issues, but several interface honesty and correctness gaps worth addressing before these samples become reference material.

📋 Key Themes & Highlights

Key Themes

• Unused parameters in tool schemas (275, 278): key in inferType and value in anonymizeValue are declared but never read. This makes the tool contract misleading — callers believe the handler uses those inputs.
• Correctness gap in CSV parsing (279): parseCSVRow silently corrupts rows containing quoted commas. At minimum, a comment should warn readers.
• Regex matches commented code (277): extractInterfaces will match interface tokens inside block comments, generating phantom type guards.
• Missing p.* intents in 276: The git hook installer tells the LLM to "write" and "chmod +x" in prose, but never declares those operations as p.write/p.bash intents — the sample doesn't demonstrate the primitives it implies.

Positive Highlights

• ✅ Good variety of addons usage: repair() on output-sensitive agents, steering() on open-ended classification.
• ✅ Subagent delegation in 274 is cleanly structured with a dedicated branchAnalyzer agent.
• ✅ p.readOptional usage in 275 is a good showcase of the optional-read primitive.
• ✅ Async defineTool handlers (276, 277) correctly use dynamic import() with node: prefix.

🧠 Reviewed using Matt Pocock's skills by Matt Pocock Skills Reviewer · sonnet46 34.9 AIC · ⌖ 4.62 AIC · ⊞ 6.3K
_{Comment /matt to run again}

[github-actions]

[/codebase-design] The value parameter is declared in the tool schema (s.unknown) but silently ignored by the handler — only fieldName and fieldType drive the output.

This creates a misleading contract: callers assume the actual value influences anonymization, but it does not. For example, a nested object value still gets a flat "[REDACTED]" string.

💡 Suggested fix

Either remove value from the parameter schema, or use it for type-aware defaults:

if (Array.isArray(value)) return [];
if (typeof value === 'object' && value !== null) return {};

[github-actions]

[/codebase-design] The key parameter is declared in inferType's schema but never used — the handler only operates on value.

Removing it from the schema keeps the interface honest and avoids confusing future readers who might expect key-based heuristics (e.g. DATABASE_URL → url even when the value is empty).

💡 Minimal fix

parameters: s.object({ value: s.string }),
handler({ value }) {

Or use key as a fallback: if value is empty but the key name contains URL, return "url".

[github-actions]

[/codebase-design] parseCSVRow splits on a bare delimiter without handling quoted fields — a value like "Smith, Jr." will be split into two cells, corrupting the table.

This is a correctness gap: the sample is named a CSV converter but doesn't handle the most basic CSV quoting rule. Either add a caveat in the description ("assumes no quoted fields"), or handle the common case.

💡 Simple quoted-field guard

handler({ line, delimiter }) {
  // naive split — does not handle quoted fields
  return { cells: line.split(delimiter).map((cell: string) => cell.trim()) };
}

For a more robust approach, use a state-machine or the existing node:readline CSV parsing.

[github-actions]

[/codebase-design] The interface-extraction regex /interface\s+(\w+)/g will match interfaces inside block comments (/* interface Foo */) and string literals, producing phantom guard functions.

For a sample intended to demonstrate real extraction this is a misleading pattern. Add a comment noting the limitation, or filter out comment lines before matching.

💡 Quick mitigation

// Strip single-line and block comments before matching
const stripped = content.replace(/\/\/.*$/gm, '').replace(/\/\*[\s\S]*?\*\//g, '');
const matches = [...stripped.matchAll(/interface\s+(\w+)/g)];

[github-actions]

[/codebase-design] The instructions say "write the script to .git/hooks/<name> using a write operation" and "run chmod +x via bash", but these are in-prompt natural-language instructions to the LLM — there are no p.write or p.bash intents to anchor them.

Without those, the agent must hallucinate file I/O rather than using declared prompt intents. This undermines the sample's value as a teaching example for rig's p.* primitives. Consider adding explicit p.write and a p.bash intent in the instructions template.

💡 Example

instructions: p`...
For each hook: ${p.write("${hook.name}", "${hook.script}")} then ${p.bash("chmod +x .git/hooks/${hook.name}")}

(Adapt to the actual rig template tag syntax for dynamic values.)