src/commands/agent.ts (1)

325-330: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Generalize the managed-section comments beyond Codex.

Gemini now also uses managed-section mode, so these comments should describe “managed-section targets” rather than Codex specifically.

Proposed wording cleanup

-   * (`[skill]`); the codex managed-section target produces ONE result whose
+   * (`[skill]`); managed-section targets produce ONE result whose
    * section aggregates every installed skill (`[...skills]`).

-  // single AGENTS.md (so every codex row shares that path — truthful, since both
+  // single root instruction file (so managed-section rows share that path — truthful, since both

Also applies to: 764-767

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/commands/agent.ts` around lines 325 - 330, The comments on the skill
result shape in agent.ts are too Codex-specific and should be generalized to
cover all managed-section targets. Update the documentation around the skills
field and the related managed-section comment block (including the matching
comment near the later section) so they refer to “managed-section targets”
instead of “Codex,” while keeping the behavior description about own-file
targets versus aggregated managed-section results.

src/commands/agent.test.ts (1)

1418-1440: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Add Gemini repeat-install coverage for the managed block.

This new suite covers append, but the PR safety claim also depends on repeated Gemini installs replacing exactly one TestSprite block while preserving surrounding GEMINI.md content. Add a second run/assertion for exactly one begin/end sentinel and unchanged user text.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/commands/agent.test.ts` around lines 1418 - 1440, The new Gemini append
test in runInstall only verifies the first install path, but it should also
cover repeat-install behavior for the managed section. Extend the existing
runInstall — gemini managed-section: append test to perform a second run with
the same target and assert that GEMINI.md still preserves the user content while
containing exactly one managed block bounded by
TARGETS.gemini.managedSection!.begin and TARGETS.gemini.managedSection!.end. Use
the existing runInstall, TARGETS.gemini.managedSection, and makeMemFs helpers to
confirm repeated installs replace the TestSprite block instead of duplicating
it.

src/lib/agent-targets.test.ts (1)

694-716: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Add a Gemini onboard rendering assertion.

The new Gemini target is installed with DEFAULT_SKILLS, but this suite only exercises Gemini with testsprite-verify. Add one test for testsprite-onboard so the shared GEMINI.md path and managed-section contribution stay covered.

Proposed test coverage

   it('renderForTarget("gemini") without body arg uses the managed-section asset', () => {
     const result = renderForTarget('gemini', 'testsprite-verify');
     expect(result.content).toContain('testsprite test run');
     expect(result.content).not.toContain('name: testsprite-verify');
     expect(result.content).not.toContain('alwaysApply:');
   });
+
+  it('renderForTarget("gemini") supports the onboard managed-section contribution', () => {
+    const result = renderForTarget('gemini', 'testsprite-onboard');
+    expect(result.path).toBe('GEMINI.md');
+    expect(result.content).toBe(ONBOARD_CODEX_LINE);
+    expect(result.content).not.toContain('---');
+  });
 });

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/lib/agent-targets.test.ts` around lines 694 - 716, Add a Gemini
onboarding coverage test in the existing renderForTarget("gemini") suite so the
shared GEMINI.md path is verified for testsprite-onboard as well as
testsprite-verify. Reuse renderForTarget and assert that the onboard target
resolves to GEMINI.md and includes the managed-section content expected from
DEFAULT_SKILLS, keeping the assertion alongside the current gemini
content-integrity tests.

src/lib/agent-targets.ts (1)

136-140: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Update pathFor docs to include Gemini’s shared path.

The comment still says only Codex lands in a shared root file, but gemini also ignores skill and returns GEMINI.md.

Proposed JSDoc tweak

- * skills coexist; the codex target always lands at the single shared `AGENTS.md`
- * (every skill's codex contribution is merged into one managed section there).
+ * skills coexist; managed-section targets land at shared root instruction files
+ * (`AGENTS.md` for codex, `GEMINI.md` for gemini), where every selected skill's
+ * contribution is merged into one managed section.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/lib/agent-targets.ts` around lines 136 - 140, The JSDoc for pathFor is
outdated: it says only codex uses a shared root file, but gemini also ignores
the skill and maps to a single shared file. Update the comment near pathFor in
agent-targets.ts to mention both codex and gemini shared landing paths, and keep
the target-specific behavior clear for own-file targets.

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

test/e2e/agent-install.e2e.test.ts (2)

146-244: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Missing gemini managed-section content-integrity test.

This suite adds a thorough codex-specific test (Lines 209–244) verifying exactly-one sentinel pair, ordering, and both skills' content in AGENTS.md. Since gemini is now a second managed-section target using its own sentinel pair (GEMINI_MANAGED_SECTION_BEGIN/END), and preserving user content around this sentinel-delimited block is a headline feature of this PR, there's no equivalent test verifying GEMINI.md's managed-section behavior (sentinel presence/count, ordering, content, and no-frontmatter-fence check).

Without this, the exact behavior described in the PR objectives — "existing content outside the TestSprite block is preserved" for GEMINI.md — is untested at the e2e level.

Suggested addition (mirrors the codex test)

it('gemini GEMINI.md contains ONE managed section with both verify and onboard content', () => {
  const tmpDir = freshTmpDir();
  runCli(['agent', 'install', '--target=gemini', '--dir', tmpDir, '--output', 'json']);

  const filePath = join(tmpDir, TARGETS.gemini.path);
  const content = readFileSync(filePath, 'utf8');

  const beginCount = (
    content.match(
      new RegExp(GEMINI_MANAGED_SECTION_BEGIN.replace(/[.*+?^${}()|[\]\\]/g, '\\$&'), 'g'),
    ) ?? []
  ).length;
  const endCount = (
    content.match(
      new RegExp(GEMINI_MANAGED_SECTION_END.replace(/[.*+?^${}()|[\]\\]/g, '\\$&'), 'g'),
    ) ?? []
  ).length;
  expect(beginCount).toBe(1);
  expect(endCount).toBe(1);
  expect(content.indexOf(GEMINI_MANAGED_SECTION_BEGIN)).toBeLessThan(
    content.indexOf(GEMINI_MANAGED_SECTION_END),
  );
  expect(content).toContain('TestSprite Verification Loop');
  expect(content).toContain('First-time setup');
  expect(content.startsWith('---')).toBe(false);
});

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@test/e2e/agent-install.e2e.test.ts` around lines 146 - 244, Add an e2e test
for the gemini managed-section target to match the existing codex coverage:
after running agent install for TARGETS.gemini, read the generated GEMINI.md and
verify exactly one GEMINI_MANAGED_SECTION_BEGIN and GEMINI_MANAGED_SECTION_END
pair exists in the correct order, the content includes both the verification and
onboard strings, and the file does not start with frontmatter fences. Mirror the
structure of the existing codex test in test/e2e/agent-install.e2e.test.ts so
the gemini-specific managed-section behavior is validated with the same
sentinel-count and content-integrity checks.

---

433-472: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Multi-target install test omits the new gemini target.

The explicit CLI arg list --target=claude,cursor,cline,antigravity,codex (Line 433) excludes gemini, even though the matrix coverage guard (Lines 822-830 in this same file) enforces that TARGETS includes gemini. This leaves the multi-target install path (targets + skills together in one invocation) untested for the very target this PR introduces.

Suggested fix

-  it('--target=claude,cursor,cline,antigravity,codex writes all targets + skills, exit 0', () => {
+  it('--target=claude,cursor,cline,antigravity,codex,gemini writes all targets + skills, exit 0', () => {
     const tmpDir = freshTmpDir();
     const result = runCli([
       'agent',
       'install',
-      '--target=claude,cursor,cline,antigravity,codex',
+      '--target=claude,cursor,cline,antigravity,codex,gemini',
       '--dir',
       tmpDir,
       '--output',
       'json',
     ]);

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@test/e2e/agent-install.e2e.test.ts` around lines 433 - 472, The multi-target
install e2e test in the agent install spec is missing coverage for the newly
added gemini target, so the combined install path is not exercised for it.
Update the target list used by the `it('--target=... writes all targets +
skills, exit 0')` case to include gemini alongside the existing targets, and
make sure the assertions in the loop naturally verify gemini’s installed output
according to its `TARGETS[gemini].mode`. Keep the test aligned with the
`TARGETS` matrix guard so this scenario covers every supported target introduced
by the CLI.

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes