refactor(#2512): propagate discovery errors and address review findings

[fullsend-ai-coder]

Change discoverAgentSlugs and loadKnownSlugs to return errors instead of swallowing them. When DiscoverRemoteAgents fails with a transient error and no usable agents are found, the error is now propagated so callers can distinguish "no harness files exist" from "discovery failed."

runUninstall and runGitHubUninstall now abort when discovery errors with no slugs, preventing silent fallback to default naming that could orphan GitHub Apps after deleting the config repo.

Also: add ADR-0045 breadcrumb comments, tighten runGitHubUninstall variable declaration to short form, rename TestDiscoverAgentSlugs_HarnessFirst to TestDiscoverAgentSlugs_HarnessDiscovery, and add transient error tests for both discoverAgentSlugs and runUninstall.

Co-Authored-By: Claude Opus 4.6 noreply@anthropic.com

---

Closes #2512

Post-script verification

• Branch is not main/master (agent/2512-refactor-config-review-findings)
• Secret scan passed (gitleaks — 619c19d688c0aa3c25af0be4fd8b07cbc8099241..HEAD)
• Pre-commit hooks passed (authoritative run on runner)
• Tests ran inside sandbox

[github-actions]

E2E tests did not run

E2E tests run automatically for org/repo members and collaborators on pull requests.

For other contributors, a maintainer must add the ok-to-test label after the latest push.

See E2E testing guide for details.

[github-actions]

Site preview

Preview: https://f35c5660-site.fullsend-ai.workers.dev

Commit: 8c1efce280858486ec46ccac600a24021cbe2923

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 4:24 PM UTC · Completed 4:36 PM UTC
Commit: 8c1efce · View workflow run →

[codecov]

Codecov Report

❌ Patch coverage is 71.42857% with 6 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
internal/cli/admin.go	69.23%	4 Missing ⚠️
internal/cli/github.go	33.33%	1 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!

[fullsend-ai-review]

Review

Findings

Medium

• [error handling inconsistency] internal/cli/admin.go:2073 — In loadKnownSlugs, when DiscoverRemoteAgents returns both an error and agents (err != nil && len(agents) > 0), but none of those agents produce valid slugs (all filtered out by role/slug validation), the function returns (nil, nil) — swallowing the error. By contrast, discoverAgentSlugs in the same scenario falls through to its bottom if err != nil check and returns the wrapped error. The two functions have inconsistent behavior for this edge case: a partial error where the only readable agents happen to be invalid (missing role or slug) will be silently swallowed by loadKnownSlugs but propagated by discoverAgentSlugs.
  Remediation: Add an if err != nil { return nil, fmt.Errorf("harness discovery failed: %w", err) } check before the final return nil, nil in loadKnownSlugs, mirroring the pattern in discoverAgentSlugs.

Low

• [documentation-comment-format] internal/cli/admin.go:2035 — The loadKnownSlugs doc comment has a stale trailing // config repo. line — a leftover from the original two-line comment that was not removed when the new multi-paragraph docstring was inserted above it.
• [redundant logging] internal/cli/admin.go:1335 — When loadKnownSlugs encounters a transient error, the user sees two warnings: one inside loadKnownSlugs ("harness discovery: ...") and one from the caller in runAppSetup ("Could not discover known slugs: harness discovery failed: ..."), producing redundant nested output.
• [error message format] internal/cli/admin.go:1611 — The fmt.Errorf in runUninstall places text after %w ("cannot discover agent slugs: %w; aborting uninstall to avoid orphaning apps"), producing an awkward triple-nested message. Functional but unusual.

---

Labels: PR modifies CLI install/uninstall slug discovery from harness wrapper files.

[ggallen]

/fs-fix. Rebase to get the latest sources. Fix so the code coverage test passes. Then address all actionalble review issues as well, even low ones.

[ggallen]

/fs-fix. Rebase to get the latest sources. Fix so the code coverage test passes. Then address all actionable review issues as well, even low ones.