fix(#6509): move runtime resolution before sandbox creation

[fullsend-ai-coder]

The backendFromConfigFile() call that validates the configured runtime (org or per-repo config.yaml) ran after sandbox.CreateWithRetry(). A typo'd runtime: value (e.g. "opencode") would create and tear down a sandbox before the validation caught the error.

Move the runtime resolution block (formerly step 5b) to new step 4a, immediately after the pre-script skip check and before sandbox creation (now step 4b). backendFromConfigFile() only reads the host filesystem config — it has no dependency on sandbox-computed values — so the move is safe.

Now an invalid runtime config fails immediately with a clear error before any sandbox is created or torn down.

Note: golangci-lint was not available in the sandbox. gofmt, go vet, and go test all pass.

---

Closes #6509

Post-script verification

• Branch is not main/master (agent/6509-move-runtime-resolution)
• Secret scan passed (gitleaks — d31168a0ce5f0d0f2c10dbbb8ab9d78c8b066167..HEAD)
• PR body secret scan: N/A (commit body path)

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 1:27 AM UTC · Completed 1:37 AM UTC

Commit: fefb410 · View workflow run →

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[fullsend-ai-review]

Looks good to me

[rh-hemartin]

/fs-fix

[fullsend-ai-coder]

🤖 Finished Fix · ✅ Success · Started 8:47 AM UTC · Completed 9:04 AM UTC

Commit: fefb410 · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $4.13

[fullsend-ai-coder]

🔧 Fix agent — iteration 1 (human-triggered)

Investigated the functional-tests CI failure triggered by /fs-fix. The code change moving runtime resolution before sandbox creation is correct — backendFromConfigFile only reads host filesystem config and has no sandbox dependencies. Code compiles, go vet passes, and all relevant unit tests pass. The CI functional test logs are inaccessible (HTTP 403), but all other recent functional-tests runs across the repo succeed, pointing to a transient infrastructure issue. No code changes made.

Disagreed (1):

1. functional-tests CI check failure: Investigated thoroughly: the code change (moving runtime resolution from step 5b to step 4a) is logically correct. backendFromConfigFile() only reads the host filesystem and has no dependency on sandbox-computed values. All variables (absFullsendDir, orgConfigPath, printer, aggMetrics) are available at the new location. The code compiles, passes go vet, and all relevant unit tests pass (TestBackendFromConfigFile_, TestPostScriptEnv_, TestPostScriptRepoEnv). The functional test failure (exit code 2) cannot be diagnosed without CI log access (logs returned HTTP 403). All other recent functional-tests runs across the repo pass successfully, suggesting a transient infrastructure issue rather than a code defect. No code changes are warranted.

Tests: passed

Next steps:

• /fs-review — request a re-review of the changes
• /fs-fix <your instruction> — run another fix pass with specific guidance
• Push commits directly — review re-runs automatically on push
  _{Updated by fullsend fix agent}

[rh-hemartin]

/fs-fix rebase and fix conflicts

[fullsend-ai-coder]

🤖 Finished Fix · ✅ Success · Started 9:32 AM UTC · Completed 9:40 AM UTC

Commit: fefb410 · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $1.59

[fullsend-ai-coder]

🔧 Fix agent — iteration 1 (human-triggered)

Rebased agent/6509-move-runtime-resolution onto origin/main. Resolved a merge conflict in internal/cli/run.go by keeping main's version — the PR's change (move runtime resolution before sandbox creation) is already incorporated in main via resolveBackendFrom() at line 954. The commit was dropped as empty. One pre-existing test failure (TestPostScriptEnv_NoSchemaAppendedWhenNoValidationLoop) exists on main and is unrelated to this PR.

Fixed (1):

1. Rebase and fix conflicts (internal/cli/run.go): Rebased branch onto origin/main, resolving a merge conflict in internal/cli/run.go. The conflict arose because main already incorporated the PR's intent (move runtime resolution before sandbox creation) via resolveBackendFrom() at line 954, well before sandbox creation at line 1466. Kept main's version in the conflict block (step 5b variable reassignment with new metric fields) and removed the PR's now-redundant step 4a block that called backendFromConfigFile() with the wrong arity. The rebase commit was dropped as empty since the PR's change is fully superseded by main.

Tests: passed

Decision points

• PR change is fully superseded by main (alternatives: Keep the empty commit for history, Drop the commit since the diff is empty; rationale: The underlying issue Move runtime resolution above sandbox creation to fail fast on invalid runtime config #6509 (runtime resolution happening after sandbox creation) was already fixed on main by resolveBackendFrom() at line 954. The PR's backendFromConfigFile() call also had the wrong arity (1 arg vs required 2). Keeping an empty commit adds noise with no value.)

---

Next steps:

• /fs-review — request a re-review of the changes
• /fs-fix <your instruction> — run another fix pass with specific guidance
• Push commits directly — review re-runs automatically on push
  _{Updated by fullsend fix agent}

[rh-hemartin]

Closing, superseeded by main.

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 10:12 AM UTC · Completed 10:25 AM UTC

Commit: fefb410 · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $4.21

[fullsend-ai-retro]

Retro: PR #6511 — move runtime resolution before sandbox creation

Timeline

1. Aug 23 01:08 — Retro agent filed issue #6509 from PR fix(#6471): validate config-driven runtimes against ValidRuntimes() #6472 human review feedback.
2. Aug 23 01:09–01:13 — Triage agent triaged, applied ready-to-code.
3. Aug 23 01:15–01:25 — Code agent created PR #6511: pure code-motion fix moving backendFromConfigFile() before sandbox creation in internal/cli/run.go (+23/−23, 1 file).
4. Aug 23 01:25–01:37 — Review agent approved ("Looks good to me", no inline comments). ready-for-merge label applied.
5. Aug 23 01:25 — CI ran: unit tests passed, E2E passed, but functional tests failed.
6. Aug 27 08:45–09:04 — Human triggered /fs-fix. Fix agent investigated (~18 min, $4.13), concluded code was correct, functional test failure was transient. Could not access CI logs (HTTP 403). No code changes.
7. Sep 2 09:30–09:40 — Human triggered /fs-fix rebase and fix conflicts. Fix agent rebased (~8 min, $1.59), found the change already incorporated on main via resolveBackendFrom(). Commit dropped as empty.
8. Sep 2 10:10 — Human closed PR as superseded by main.
9. Sep 2 10:12–10:16 — Human triggered /fs-triage on issue Move runtime resolution above sandbox creation to fail fast on invalid runtime config #6509. Triage agent verified fix on main ($0.77), closed issue.

Total agent cost for zero merged code: ~$6.49 across fix and triage agents alone (code and review agent costs not reported in comments).

What went well

• The retro-to-triage-to-code pipeline worked smoothly: issue Move runtime resolution above sandbox creation to fail fast on invalid runtime config #6509 was filed, triaged, and a correct PR opened within 17 minutes.
• The code agent's fix was correct — it accurately identified that backendFromConfigFile() has no dependency on sandbox-computed values.
• The fix agent's second run correctly identified the PR was superseded and dropped the empty commit rather than forcing a no-op push.
• The triage agent efficiently verified the fix was on main and closed the issue.

Evidence for existing issues (skipped as proposals)

• #1424 (ready-for-merge without CI check): This PR received ready-for-merge despite functional tests failing. The label sat on the PR for 10 days creating a misleading signal. This is another data point confirming the issue's impact.
• #3140 and related (CI log access): Fix agent hit HTTP 403 trying to read functional test logs, forcing it to reason indirectly about the failure.
• #2597 / #3419 (fix agent zero-change cost): First fix run cost $4.13 with no code changes, only investigation.
• Review depth (#6891, #6322): Review agent approved a code-motion change with an empty-body APPROVED review and no inline comments. The change was simple enough that this was likely fine, but the review provided no evidence of having verified the safety of the code motion (e.g., checking that backendFromConfigFile doesn't depend on sandbox-computed values).

Proposals filed

• Periodic rebase check to auto-close agent PRs superseded by main (in fullsend-ai/fullsend)