test(workflows): name a step that runs an unknown script or subcommand

[Vivswan]

Before

step run: bun .github/scripts/release-pipeline.ts package-commits
wiringProblems(post-green.yml) -> ""

step run: bun .github/scripts/post-gren-steps.ts probe
wiringProblems(post-green.yml) -> six downstream faults ("reads steps.token.outputs.proceed, which no
                                  earlier step writes" ...), none naming the script

After

build: "Package this commit under its build tag, prune the window, and point latest at the newest main source" runs release-pipeline package-commits, a subcommand .github/scripts/release-pipeline.ts does not dispatch

build: "Check the token can push" runs .github/scripts/post-gren-steps.ts, a script this relation does not know

How

• One registry of steps scripts in the test: each script's exported COMMANDS table (what it dispatches) beside its STEP_OUTPUTS (what those write).
• A script step naming a script or subcommand outside it is a wiring fault naming the step.
• A broad detector and a strict reader: any line handing bun a steps script is detected; a detected step the reader cannot parse fails the control.
• post-green-steps exports its table, as release-pipeline already does.
• CONTRIBUTING.md names release-pipeline/tags.ts as where the git topology lives, where the split moved it.

Proof

• Red then green: the typo rows failed against the old relation (received "" and the six-line list above), then passed; the doubled-space row went red again under a single-space regex.
• Tests: test/workflows 84 green (4 new rows), test/scripts/entry.test.ts 4 green; typecheck, biome, knip green.

Technical details

• Built on refactor(scripts): dispatch subcommands and run mains through one lib #523: the known-subcommand set is the handler table that PR introduced; STEP_OUTPUTS alone lists only the subcommands that write outputs (release-pipeline declares 1 of 11), so it cannot judge a subcommand.
• Steps the wider regex newly reads: the GITHUB_SHA="$SOURCE_SHA" bun ... package-commit|package|retag-major|npm-publish next|npm-publish stable steps; their declared outputs are unchanged (none), so isProbe and the gate chain read as before.
• Inherited keys are not entries: the registry is a Map and the outputs lookup is an own-key read, so constructor.ts and a constructor subcommand are named as unknown instead of throwing in outputsWritten; a row pins each.
• One more cross-file pin: every subcommand a script declares outputs for is one it dispatches.
• Shapes the reader refuses, loudly: a newline between tokens, a bun flag, bun run, a line continuation, a prefix bash reads as a command (1PAT_SET="true"), single-quoted assignments, multi-command blocks; each is detected and fails the control rather than passing unjudged, and the control pins the first five. Any run text with bun ahead of .github/scripts/ is detected.
• Library: none applies; the relation is a regex over yaml the test already parses.
• Files: test/workflows/post-green-workflow.test.ts, .github/scripts/post-green-steps.ts (the export), CONTRIBUTING.md (line 60).
• Accounting: 3 files, +129 / -22.

[Copilot]

Copilot review overview

🟢 Approval recommended

The validation is consistent with the dispatcher contract and covers the relevant failure modes.

Review effort: Balanced
Findings: None

What changed in this PR

Adds workflow wiring checks that identify unknown scripts and subcommands before post-green hooks run.

Changes:

• Exports the post-green command registry.
• Validates workflow script invocations against registered commands.
• Corrects the release topology documentation path.

File	Description
.github/​scripts/​post-green-steps.ts	Exports and reuses the command table.
test/​workflows/​post-green-workflow.test.ts	Adds script parsing, validation, and negative controls.
CONTRIBUTING.md	Updates the topology source path.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Copilot review overview

🟢 Approval recommended

The focused test and documentation changes consistently implement the described workflow validation.

Review effort: Balanced
Findings: None