Skip to content

test(antigravity): realign mapping assertions with the pruned reference#1992

Closed
cwest wants to merge 1 commit into
obra:devfrom
cwest:topic/antigravity-tool-mapping
Closed

test(antigravity): realign mapping assertions with the pruned reference#1992
cwest wants to merge 1 commit into
obra:devfrom
cwest:topic/antigravity-tool-mapping

Conversation

@cwest

@cwest cwest commented Jul 15, 2026

Copy link
Copy Markdown

This PR targets the dev branch.

Who is submitting this PR? (required)

Field Value
Your model + version Claude Opus 4.8 (claude-opus-4-8), via Google Vertex AI
Harness + version An internal Google agent harness (not publicly named)
All plugins installed None relevant to this change; the superpowers plugin was installed on the real Antigravity CLI (agy 1.1.1) for live verification
Human partner who reviewed this diff Casey West (@cwest) — reviewed the complete diff and verified it end-to-end, including a live run on the real agy CLI

What problem are you trying to solve?

tests/antigravity/test-antigravity-tools.sh fails on dev:

$ bash tests/antigravity/test-antigravity-tools.sh
test-antigravity-tools: checking Antigravity tool mapping
FAIL: mapping does not document view_file as the file/skill-read tool   # exit 1

e7ddc25 (#1847) intentionally trimmed the skill-loading explainer and the generic action→tool table out of antigravity-tools.md (pruning to "the harness-specific notes that still carry weight"), but did not update tests/. The test still greps for tokens the prune removed — view_file, IsSkillFile, SKILL.md, run_command, grep_search — so the antigravity suite has been red ever since. The doc is correct as-is; the test is stale.

What does this PR change?

Realigns tests/antigravity/test-antigravity-tools.sh with the slimmed reference: it now asserts only what the doc still documents and agents rely on — subagent dispatch (invoke_subagent, self/research), the write_to_file/replace_file_content tools, and the task-artifact mechanism — plus the SKILL.md Platform Adaptation link. Drops the assertions for the intentionally-removed tokens. One file (test only); the reference doc is untouched.

Is this change appropriate for the core library?

Yes — it repairs an existing core test so the antigravity suite is green again. No new skill, domain, or third-party integration; no product behavior change.

What alternatives did you consider?

  1. Restore the pruned content to antigravity-tools.md (turn the test green by re-adding view_file/IsSkillFile/etc.). Rejected: Prune per-harness tool-mapping boilerplate #1847 removed that content deliberately, and a clean agy run confirms agents don't need it (skills load and brainstorming auto-triggers without it). Re-adding it would reintroduce exactly what the prune set out to remove.
  2. Realign the test with the slimmed doc (this PR). Chosen — it matches the intent of Prune per-harness tool-mapping boilerplate #1847 and keeps the doc lean.

Does this PR contain multiple unrelated changes?

No. Single file, single concern: make the stale antigravity test match the intentionally-pruned reference.

Existing PRs

#1987 (open) proposes the same realignment across two files (tests/antigravity + tests/pi). This PR is the narrower, antigravity-only version — happy to defer to #1987 if you'd rather take the combined change; flagging it so you can choose.

#1902 (closed) proposed the opposite fix (restoring the doc). Per the reasoning above, the doc-restore direction re-adds content #1847 intentionally removed, so this PR fixes the test instead.

Environment tested

Harness Harness version Model Model version/ID
Antigravity CLI (agy) 1.1.1 Antigravity default (Gemini) not surfaced by agy

Repo suite (no model): bash tests/antigravity/run-tests.sh → PASS; bash tests/hooks/test-session-start.sh → PASS; git diff --check → clean.

New harness support (required if this PR adds a new harness)

N/A — no new harness. As corroboration that the removed details aren't needed at runtime: on a clean, isolated agy 1.1.1 profile (agy plugin install <repo>) the plugin loaded 14 skills + the SessionStart hook, and a live session auto-triggered brainstorming before any code was written — without the pruned mapping lines present.

Evaluation

  • Before: the antigravity suite fails on dev with the error above.
  • After: bash tests/antigravity/run-tests.sh passes; the test still catches real regressions (removing invoke_subagent, the self/research types, or the task-artifact mechanism from the doc still fails it).
  • A live agy run confirmed skills load/trigger without the removed tokens — i.e. the doc is correct as slimmed.

Rigor

  • N/A — not a skills change (this is a test-harness assertion realignment)
  • This change was tested adversarially, not just on the happy path (full antigravity suite + hooks test + whitespace check + live agy run; confirmed the trimmed test still fails when a still-documented token is removed)
  • I did not modify carefully-tuned content (Red Flags table, rationalizations, "human partner" language)

Human review

  • A human (Casey West, @cwest) has reviewed the COMPLETE proposed diff before submission

@cwest
cwest marked this pull request as ready for review July 15, 2026 16:52
@cwest
cwest marked this pull request as draft July 15, 2026 17:02
e7ddc25 (obra#1847) intentionally trimmed the skill-loading explainer and the
generic action→tool table from antigravity-tools.md, pruning to "the
harness-specific notes that still carry weight." But tests/antigravity still
asserted the removed tokens (view_file, IsSkillFile, SKILL.md, run_command,
grep_search), so the antigravity suite has been red on dev ever since.

Realign the test with the slimmed reference: assert only what the doc still
documents — subagent dispatch (invoke_subagent, self/research types), the
write_to_file/replace_file_content tools, and the task artifact mechanism —
plus the SKILL.md Platform Adaptation link. Drop the assertions for the
intentionally-removed tokens.

Agents don't need the removed details: verified with a clean agy run where the
plugin installs, skills load, and brainstorming auto-triggers without them.

Related: obra#1987 independently proposed realigning these assertions (antigravity
+ pi); this is the narrower antigravity-only change.
@cwest
cwest force-pushed the topic/antigravity-tool-mapping branch from c83e327 to cc4e345 Compare July 15, 2026 17:08
@cwest cwest changed the title docs: restore Antigravity tool-mapping reference (fixes failing antigravity suite) test(antigravity): realign mapping assertions with the pruned reference Jul 15, 2026
@cwest

cwest commented Jul 15, 2026

Copy link
Copy Markdown
Author

Closing in favor of #1987 — it makes the same test realignment and also covers
the pi test (this one was antigravity-only), so it's the more complete fix.
I verified it and left details over there.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant