test(pi): trim stale tool-mapping assertions

[muunkky]

Who is submitting this PR? (required)

Field	Value
Your model + version	Claude Opus 4.8 (claude-opus-4-8), 1M context
Harness + version	Claude Code 2.1.207
All plugins installed	gitban (muunkky.github.io/gitban-site), frontend-design, skill-creator, claude-patent-creator-standalone. Not superpowers itself.
Human partner who reviewed this diff	Cameron Rout (@muunkky)

What problem are you trying to solve?

tests/pi/test-pi-extension.mjs fails on dev:

$ node --test tests/pi/test-pi-extension.mjs
not ok 6 - pi tools reference documents pi-specific mappings
  name: 'AssertionError'
# pass 5
# fail 1

The test loops over ['Skill', 'Task', 'TodoWrite', 'read', 'write', 'edit', 'bash'] and asserts each matches pi-tools.md. write, edit and bash are gone — e7ddc25 ("Prune per-harness tool-mapping boilerplate") removed those generic mapping rows on purpose. The test was never updated, so it fails on write.

What does this PR change?

The assertion in that one test. It now checks the mapping table itself:

-  for (const expected of ['Skill', 'Task', 'TodoWrite', 'read', 'write', 'edit', 'bash']) {
-    assert.match(text, new RegExp(expected));
-  }
+  const rows = text.split('\n').filter((line) => line.startsWith('|'));
+  assert.ok(rows.some((row) => /subagent/i.test(row)), 'mapping table documents subagent dispatch');
+  assert.ok(rows.some((row) => /todo|task/i.test(row)), 'mapping table documents task tracking');

pi-tools.md is not touched — the prune stays pruned. git diff --numstat dev → 3 3 tests/pi/test-pi-extension.mjs. Suite goes 5 pass / 1 fail → 6 pass / 0 fail.

Is this change appropriate for the core library?

Yes — it repairs a core test that is currently red.

What alternatives did you consider?

Restoring the deleted rows to pi-tools.md would also turn the test green, and that's what the earlier attempt did. It's the wrong fix: it reverts a deliberate decision to satisfy a stale assertion.

A straight trim of the token list — my first attempt, and it was wrong for a reason worth stating. I trimmed the list to ['pi-subagents', 'subagent', 'TODO.md', 'TodoWrite'] and it passed. Then I deleted the entire mapping table from pi-tools.md and it still passed: all four tokens survive in the prose around the table. The trimmed list guarded nothing — the same defect the original list has. So the assertions here anchor on the table rows instead. I verified it by deleting the table and watching the test fail, then restoring it and watching it pass.

That's three added lines rather than one, which is more than a pure trim. If you'd rather have the one-line version, it's yours, but it doesn't test anything.

Does this PR contain multiple unrelated changes?

No. One file.

Existing PRs

• I have reviewed all open AND closed PRs for duplicates or prior art
• Related PRs: Document Pi mappings for legacy skill vocabulary #1903 (closed)

#1903 (@tianma-if) found this failure and the credit for spotting it is theirs. It was closed partly as one of a ten-PR batch, but also on the merits: it restored the pruned Read/Write/Edit/Bash rows, reversing e7ddc25. This does the opposite and leaves pi-tools.md alone. I said on that thread I'd stand aside if they wanted to write it; there's been no reply.

Environment tested

Harness	Harness version	Model	Model version/ID
Claude Code	2.1.207	Claude Opus 4.8	claude-opus-4-8 (1M context)

Reproduced the failure on a clean dev checkout (Linux, node --test) before changing anything; confirmed green after; and confirmed the new assertions fail when the mapping table is removed.

Evaluation

Not applicable — test-only change, no skill prose touched, so there's no agent behaviour to eval. The verification is the test: red before, green after, and red again if the thing it checks is deleted.

Rigor

• If this is a skills change: I used superpowers:writing-skills and completed adversarial pressure testing
• This change was tested adversarially, not just on the happy path
• I did not modify carefully-tuned content (Red Flags table, rationalizations, "human partner" language) without extensive evals

First box unticked: this is test-only, no skill content changed, so writing-skills doesn't apply.

The adversarial test is the mutation above: I deleted the mapping table and confirmed the assertions fail. That's what caught my first attempt, which passed with the table gone.

Human review

• A human has reviewed the COMPLETE proposed diff before submission

[muunkky]

Superseded by #1987, which fixes this test and the antigravity one. @gaurav0107 took the table-scoped assertion in 0873545 — verified it now fails on a pi-tools.md with the mapping table stripped and passes on dev, which is the regression the old token list was missing.