fix(agent-lifecycle): stop showing credentials_injection: failed for subscription agents (#612)

[dolho]

Closes #612

Summary

Subscription-mode agents authenticate via CLAUDE_CODE_OAUTH_TOKEN env var (SUB-002) and have no .credentials.enc file. The agent-start path ran inject_assigned_credentials unconditionally, the encryption service raised ValueError("No .credentials.enc file found in agent workspace"), and the caller's "not found" in str(e).lower() substring check missed it (the actual message contains "file found", not "not found").

The misleading {"credentials_injection": "failed"} leaks to the agent-start response and prompts operators to re-assign subscriptions or recreate containers — disrupting running state for no reason. We hit this exact false alarm earlier today restarting agent-trinity-system after a subscription assignment.

Fix

Two-part:

1. CredentialsFileNotFoundError (ValueError subclass) in services/credential_encryption.py. import_to_agent raises this specific type when .credentials.enc is absent, so callers can catch it without fragile substring matching. Subclassing ValueError preserves the admin import endpoint's existing 400 behavior — that path should surface a real error when the operator explicitly asks for import.

2. Subscription-mode short-circuit in inject_assigned_credentials. db.get_agent_subscription_id() check fires before the import path runs; returns {"status": "skipped", "reason": "subscription_mode", "detail": "agent authenticates via CLAUDE_CODE_OAUTH_TOKEN; file-based credential injection is not used"} with explicit mention of the env-var path so operators see this is by design.

The substring-match path is removed entirely — the explicit exception subclass replaces it.

Why two changes

Either alone would close the cosmetic bug, but together they're load-bearing:

• Subscription-only fix: solves the reported case but leaves the substring-match landmine for any future error message change.
• Exception-subclass-only fix: subscription agents would still call import_to_agent (a no-op for them) and waste retry budget on the doomed path.

Together they short-circuit the common case AND make the rare case (subscription agent that also has a .credentials.enc for non-Anthropic creds) report cleanly if it ever fails.

Tests

tests/unit/test_inject_assigned_credentials.py — 8 cases:

• subscription mode short-circuits without consulting the service
• file-missing returns skipped (not failed)
• CredentialsFileNotFoundError remains a ValueError (admin path compat)
• happy path returns success with file list
• empty decrypted dict returns skipped
• other ValueError shapes retry then surface as failed
• transient ValueError recovers on retry
• missing CREDENTIAL_ENCRYPTION_KEY returns skipped

$ python -m pytest tests/unit/test_start_agent_skip_inject.py tests/unit/test_inject_assigned_credentials.py tests/unit/test_credential_inject_allowlist.py -q
32 passed in 121.02s

Test plan

• Unit tests pass (32/32 incl. existing bug: credential injection runs on health-check recovery path against already-running containers #421 skip-inject tests)
• Compile-check passes
• Live verification — blocked by unrelated REDIS_URL config drift on this dev branch (SEC: Redis has no requirepass — agent containers have unauthenticated full read/write access to platform task queue (AISEC-C1) #589 not yet propagated to local .env)
• Manual verification post-merge: stop and start a subscription-mode agent; expect {"credentials_injection": "skipped", "credentials_result": {"reason": "subscription_mode", ...}}

🤖 Generated with Claude Code

[vybe]

Clean two-part fix — exception subclass removes the fragile substring match and the subscription short-circuit fires before the service is even consulted. 8 unit tests cover all the edge cases. LGTM.