[rig-tasks] Daily rig evaluation — 2026-08-09 — 10/10 passed

[github-actions]

Summary

Task	Description	Typecheck	Key finding
1 (reused)	TS dead export finder	✅ pass	Initial as const on array literal caused TS1355 — replaced with plain return; otherwise solid p.glob + async defineTool pattern
2 (reused)	Pkg scripts documenter	✅ pass	Unused name param in handler caused TS6133 — destructured only command; p.read + p.write + s.enum category pattern works well
3 (reused)	Git diff stats summarizer	✅ pass	p.bash + sync defineTool + s.optional clean combination; s.enum for category
4 (reused)	Dotenv template generator	✅ pass	p.readOptional + p.glob + async defineTool + p.write — all four intent types in one agent, repair addon
5 (reused)	TS class hierarchy extractor	✅ pass	async defineTool + steering addon; s.record + s.optional(s.string) output structure
6 (reused)	Git bisect helper	✅ pass	steering addon with maxTurns:8 + sync defineTool selectMidpoint; s.enum confidence
7 (new)	Shell shebang validator	✅ pass	p.glob + async defineTool + s.record output with s.optional(s.string) for shebangLine; repair addon
8 (new)	Git log graph summarizer	✅ pass	p.bash + sync defineTool with s.enum return; steering addon
9 (new)	TS complexity scorer	✅ pass	p.glob + async defineTool computing s.number score + s.enum complexity; repair addon
10 (new)	Parallel multi-tool workflow	✅ pass	workflow() with Promise.all of two call() invocations + call.json for coordination; required two fixes

---

Problems encountered

Task 1 — TS Dead Export Finder

as const applied to an array returned from an async handler caused TS1355 ("A 'const' assertion can only be applied to references to enum members, or string, number, boolean, array, or object literals"). Plain return works correctly.

Task 2 — Pkg Scripts Documenter

Unused parameter name in handler destructuring caused TS6133. Fixed by removing it from destructuring.

Task 10 — Parallel Multi-Tool Workflow

Two separate issues:

1. workflow() does not accept an output field — the body return type is inferred from TypeScript. Removed output from the spec.
2. parallel() from the workflow body context enforces uniform return types: using parallel([() => call(agentA, ...), () => call(agentB, ...)]) where A and B have different output schemas caused a type mismatch. Fixed by using Promise.all instead, which correctly resolves each typed call independently.

---

Improvement opportunities

Missing or undiscoverable schema helpers (s.*)

• s.int vs s.number distinction is well-documented but generated code initially used s.number for count fields. A lint rule prompting s.int for integer counts would help.

Missing or undiscoverable prompt helpers (p.*)

• No issues; p.glob, p.bash, p.read, p.readOptional, p.write all used correctly across the 10 programs.

Error message quality

• TS2559 ("Type has no properties in common") for workflow({ output: ... }) is opaque. The error body mentions WorkflowSpec but doesn't say "output is not a valid workflow field". A custom diagnostic or lint rule saying "workflow() does not support an output field; the body return type is inferred" would save time.
• TS2322 from parallel() with heterogeneous agent outputs is confusing — the error points to the second element of the tuple as mismatched. It is not obvious that parallel unifies types vs Promise.all which preserves each independently.

API ergonomics

• parallel() in workflow bodies enforces a single result type across all thunks, which is surprising when using typed subagents with different output schemas. Promise.all works instead but loses the concurrency limiter integration. A parallel overload that accepts a typed tuple [() => Promise<A|null>, () => Promise<B|null>] and returns [A|null, B|null] would greatly improve ergonomics for heterogeneous parallel calls.

Candidate lint rules

Rule: no-workflow-output-field

• Invalid: workflow({ meta, output: s.object({...}), body })
• Valid: workflow({ meta, body }) — body return type is inferred
• Why confusing: agent() has an output field, so models try the same for workflow(). The field is silently dropped (or causes a type error).
• Autofix: remove the output property from the workflow() call.

Rule: prefer-s-int-for-counts

• Invalid: totalFiles: s.number
• Valid: totalFiles: s.int
• Why confusing: s.number technically works but loses the integer semantic contract. Models use it for count fields.
• Autofix: replace s.number with s.int when the field name ends in Count, Total, Lines, Steps, etc.

Documentation gaps

• SKILL.md composition invariants mention that workflow() fields include meta, input, body but do not explicitly call out that output is absent (unlike agent()). Adding a one-liner "No output field — body return type is inferred" would prevent a common mistake.
• parallel() type constraint behavior (uniform vs tuple) is not described in SKILL.md or the dynamic-workflows reference. Worth adding a note that heterogeneous parallel calls should use Promise.all instead.

---

Tasks run today

• (reused) TypeScript dead export finder — p.glob + async defineTool + steering + repair
• (reused) package.json scripts documenter — p.read + defineTool + p.write + repair
• (reused) Git diff stats summarizer — p.bash + defineTool + s.enum + repair
• (reused) Dotenv template generator — p.readOptional + p.glob + async defineTool + p.write + repair
• (reused) Multi-file TypeScript class hierarchy extractor — p.bash + async defineTool + steering
• (reused) Git bisect helper — p.bash + defineTool + steering maxTurns:8
• (new) Shell script glob validator — p.glob + async defineTool + repair + s.record
• (new) Git log graph summarizer — p.bash + sync defineTool + steering + s.enum
• (new) TypeScript complexity scorer — p.glob + async defineTool + s.number + repair
• (new) Workflow parallel multi-tool tester — workflow + Promise.all + call.json coordination

Generated by Daily Rig Task Generator · sonnet46 105.5 AIC · ⌖ 9.34 AIC · ⊞ 6.8K · ◷