ARCH-009: manifest.py completion checks exclude completed_with_failures status

[Muizzkolapo]

Summary

Two methods in workflow/managers/manifest.py check for exact "completed" status:

is_action_completed() (L221-224):

def is_action_completed(self, action_name: str) -> bool:
    action = self.manifest.get("actions", {}).get(action_name)
    return action is not None and action.get("status") == "completed"

get_completed_actions() (L341-347):

return [name for name, info in actions.items() if info.get("status") == "completed"]

Both exclude completed_with_failures. Any manifest logic that relies on these methods (e.g., determining which actions have run, resume logic, reporting) will treat partially-failed actions as incomplete.

Impact

• Actions that completed with partial item failures are invisible to manifest queries
• Resume scenarios may re-run actions that already produced valid output
• Reporting undercounts completed work

Suggested Fix

def is_action_completed(self, action_name: str) -> bool:
    action = self.manifest.get("actions", {}).get(action_name)
    return action is not None and action.get("status") in {"completed", "completed_with_failures"}

Same pattern for get_completed_actions().

Note: If callers need to distinguish clean vs partial completion, add a separate is_action_completed_clean() or accept a strict parameter. Don't break the common case.

Related

• ARCH-007 (ARCH-007: Batch completion path never detects partial item failures (completed_with_failures is online-only) #89) — batch never sets this status (masks the bug for batch actions)
• ARCH-004 (ARCH-004: Formalize action state machine to prevent mode-specific state leaks #86) — formalized state machine would define terminal/complete sets once

Files Affected

• agent_actions/workflow/managers/manifest.py — L221-224, L341-347

Acceptance Criteria

• is_action_completed() returns True for both completed and completed_with_failures
• get_completed_actions() includes actions with completed_with_failures
• Callers that need strict "no failures" check have an explicit method for it

[Muizzkolapo]

Heads-up — bundle with Phase 2 PR

Same pattern as #90. Two methods in manifest.py need the set check:

# L221-224:
return action is not None and action.get("status") in {"completed", "completed_with_failures"}

# L341-347:
return [name for name, info in actions.items()
        if info.get("status") in {"completed", "completed_with_failures"}]

Without this, resume logic will re-run actions that already produced valid (partial) output.

[Muizzkolapo]

Fixed in PR #92 — both is_action_completed() and get_completed_actions() in manifest.py now use set checks.