new-input

name	new-input
description	Generate committed example inputs by profiling real workloads: workload scripts, generation, inspection, and validation. Use when asked to add an input or example for a language, profiler, or configuration.
argument-hint	[language, origin, format, or guidance]
allowed-tools	Bash(pnpm generate-inputs *)

Generate committed example inputs in examples/input/ by profiling real workloads.

Input

$ARGUMENTS

Workflow

If loaded from /new-format, skip step 1 and return to that workflow after step 3: the remaining steps assume the converter already exists.

Check support

1. Determine each input's language, origin, and format, then check they're supported:
   • Format not registered in src/formats/registry.ts: STOP, load /new-format, and follow its workflow instead; it implements the format end-to-end and returns here to generate inputs
   • Origin not registered in src/origins/specs/index.ts: load /new-origin and follow its workflow first; input filenames carry the origin's registered ID, and every committed input must resolve to it

Generate

2. Add or update workload scripts in scripts/inputs/, one per emitting language (<lang>.sh, sharing the _common.sh helpers), then run pnpm generate-inputs <langs>:

   • Profile a real, popular program with realistic input. NEVER a toy workload or synthetic input
   • Native-only tools run in Docker via docker_capture; host tools may need additions to scripts/inputs/flake.nix
   • Name outputs <lang>.<origin>.<config?>.<base|current>.<ext>; the second token must be the origin's registered ID, and the extension must match the converter's extension
   • When a profiler can export multiple supported formats, export all from a single recording
   • Keep every input under GitHub's 100 MB limit (verify_generated_input fails oversized captures). An oversized input means the workload does too much work while recording (high-volume configs like nativemem record every event): shrink the workload's input or iteration count, not the profiler flags, so the capture stays realistic

3. Inspect the generated bytes by hand: confirm each file is the intended format, the workload's own code dominates the samples, and the origin's evidence (marker entries or a parser origin hint) is present

Validate

4. Convert each input with node src/cli/index.ts <input>, and each base/current pair as a diff. Sanity-check the output against the workload

5. Run the detected-input-origins test (pnpm test src/origins/index.test.ts): every committed input must resolve to the origin in its filename. NEVER commit an input that resolves elsewhere: give the origin a marker or a parser origin hint, or make the workload realistic enough to carry the origin's evidence

Document and finish

6. For a language we don't list yet: add a languageMetas entry (name, aliases, extensions) in src/cli/languages.ts and generation instructions in docs/languages/<language>.md, then verify the docs render with node src/cli/index.ts --help <language>. If a new config token needs a display label beyond title-casing, add it to configNames in src/cli/examples.ts

7. pnpm update-examples, then pnpm update-readme