fix(#5241): add missing --file flag to ImportProfile openshell command

[fullsend-ai-coder]

Summary

• Fix: Add missing --file flag to ImportProfile when invoking openshell provider profile import. The profile path was passed as a bare positional argument, but OpenShell requires --file <FILE> or --from <FROM>.
• Test: Add TestImportProfile_UsesFileFlag that uses a fake openshell script to verify the --file flag is present in the import command args.

Related Issue

Closes #5241

Changes

• internal/sandbox/sandbox.go: Add "--file" before profilePath in the exec.CommandContext call inside ImportProfile
• internal/sandbox/sandbox_test.go: Add TestImportProfile_UsesFileFlag test that captures openshell args and asserts --file is included

Testing

• go test ./internal/sandbox/ — all 60 tests pass
• go vet ./internal/sandbox/ — clean
• scan-secrets --staged — no leaks found
• The sibling ImportProfiles function already uses --from dir correctly and is not affected

Checklist

• Minimal change — one-line fix plus one test
• Follows existing test patterns (fake openshell script)
• No secrets or sensitive data committed

🤖 Generated with Claude Code

---

Closes #5241

Post-script verification

• Branch is not main/master (agent/5241-fix-importprofile-file-flag)
• Secret scan passed (gitleaks — 6a3bfb93a6275f984b2892798edae2dfdb93b56b..HEAD)
• PR body secret scan passed (gitleaks — no-git)
• Pre-commit hooks passed (authoritative run on runner)
• Tests ran inside sandbox

[github-actions]

Site preview

Preview: https://d6768d1d-site.fullsend-ai.workers.dev

Commit: 7b4ddaefa04aa14aa1a7443468e0806e8a07442b

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[maruiz93]

/fs-review

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 2:45 PM UTC · Completed 2:52 PM UTC
Commit: cc7a526 · View workflow run →

[fullsend-ai-review]

Looks good to me

---

Labels: PR modifies internal/sandbox/ — sandbox component bug fix

[waynesun09]

Ran 3 independent reviews (Claude x2, Grok). All clean, no findings that block this.

• Fix matches the exact error string from #5241 (`<--file |--from >`).
• No other `openshell` call site in the codebase has the same positional-arg-instead-of-flag bug (grepped all provider/profile invocations — create/update/delete all correctly use flags or documented positional signatures already).
• New test is a genuine regression guard, not trivially-passing — traced manually: pre-fix, the logged args would be `provider profile import /some/my-profile.yaml` with no `--file`, which would correctly fail the assertion.
• No injection surface added (argv form throughout, `--file` is a hardcoded literal) — and as a minor bonus, explicitly binding the path to `--file` is actually more robust than the old bare positional arg, which could have been misparsed if a path ever began with `-`.
• Test isolation confirmed safe: no `t.Parallel()` in this file, `t.Setenv` auto-restores PATH, and the fake script's `$3` check correctly targets only the `import` invocation, not the preceding `delete` call.

One important process note, not a code issue: merging this alone won't turn fullsend-ai/agents#211's CI green. Confirmed agents#211's functional-tests resolves FULLSEND_VERSION via `gh release view --repo fullsend-ai/fullsend` (floating "latest"), not a source build — and the current latest release (v0.31.0, published 2026-07-16) predates this fix. A new release tag needs to be cut after this merges before #211's CI will actually pick it up.

Approving.

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 3:51 PM UTC · Completed 4:07 PM UTC
Commit: 7b4ddae · View workflow run →

[fullsend-ai-retro]

Retro: PR #5245 — fix(#5241): add missing --file flag to ImportProfile

Timeline

Time (UTC)	Event
Jul 17, 14:14	Issue #5241 filed by maruiz93 — well-specified bug with exact line, fix, and reproducer
Jul 17, 14:15	Triage agent starts
Jul 17, 14:18	Triage completes (3m 44s, $0.55) — labels: type/bug, priority/high, component/sandbox, ready-to-code
Jul 17, 14:19	Code agent starts
Jul 17, 14:26	PR #5245 created (6m 59s, $1.76) — 1-line fix + regression test, 41 tool calls
Jul 17, 14:29	CI passes (all 60 tests green, coverage clean)
Jul 17, 14:44	maruiz93 triggers /fs-review
Jul 17, 14:45	Review agent starts
Jul 17, 14:53	Review agent approves (8m 19s, $2.03) — 3 sub-agents, zero findings, posts "Looks good to me"
Jul 18, 15:37	waynesun09 approves with detailed confirmatory analysis
Jul 18, 15:49	PR merged

Total pipeline cost: ~$4.34 (triage + code + review). Rework: zero — correct fix on first try.

What went well

• Exceptional issue quality → fast resolution. The issue included exact line number, exact fix code, reproducer, and impact analysis. This enabled the code agent to implement the correct fix in under 3 minutes of active time with zero exploration waste.
• Code agent efficiency. 41 tool calls, no dead-end file reads, clean workflow: read issue → read source → apply fix → add test → verify → commit. The $1.76 cost is proportionate to the work.
• Correct approval outcome. Both the review agent and human reviewer approved, and the fix was indeed correct.

Improvement opportunities

Two proposals filed — see linked issues below.

1. Review agent APPROVE comments lack confirmatory reasoning. The review agent posted "Looks good to me" with no substantive commentary. Meanwhile, the human reviewer provided five specific verifications: confirmed the fix matches the error string, grepped all openshell call sites for the same bug pattern, traced pre-fix behavior to verify the test isn't trivially passing, confirmed no injection surface, and verified test isolation safety. The review agent's sub-agents performed similar analysis internally but none of it was surfaced.

2. Code agent wastes tokens fighting pre-commit hooks in sandbox. The agent spent 3 commit attempts (~$0.15-0.20 and ~30s) diagnosing and working around pre-commit hook failures caused by the L7 proxy blocking git-upload-pack. Since hooks run authoritatively in the post-script outside the sandbox, this in-sandbox friction is pure waste.

Proposals filed

• Review agent should surface confirmatory reasoning in APPROVE comments, not just "Looks good to me" (in fullsend-ai/agents)
• Code agent should pre-disable git hooks inside sandbox to avoid wasted commit attempts (in fullsend-ai/agents)