fix(agents): raise tool-call budget to 300 (halt) / 100 (warn) (#464)

* fix(agents): raise the tool-call budget — 150 halted legitimate work mid-task

Repeated budget-sweep bounces: a dev at the 150 ceiling gets its
container killed seconds after a real commit, burning a spawn and the
resumed agent's re-verification turns. 300 keeps the runaway guard while
clearing a real task's footprint; warn scales to 100.

* chore(docs): reflow hard-wrapped prose from the #401 merge

* chore(foundation): regenerate lifecycle artifacts; reflow inherited prose

---------

Co-authored-by: Renn F <rennf93@users.noreply.github.com>

roboco/foundation/policy/agent_loop.py

@@ -36,8 +36,12 @@ class BudgetPolicy:
     defaults). The dataclass holds the canonical defaults.
     """
 
-    tool_call_warn_at: int = 50
-    tool_call_halt_at: int = 150
+    # 150 interrupted legitimate multi-file dev work mid-task (repeated
+    # budget-sweep bounces — each a wasted spawn plus the resumed agent's
+    # re-verification turns); 300 keeps the cost guard while clearing a
+    # real task's footprint. Warn scales with it.
+    tool_call_warn_at: int = 100
+    tool_call_halt_at: int = 300
     loop_threshold: int = 3  # same tool+args repeats to trigger
     loop_window: int = 10  # rolling-window size
     loop_action: Literal["warn", "halt"] = "halt"  # NEW: was effectively "warn"

tests/foundation/test_agent_loop.py

@@ -11,8 +11,8 @@ from roboco.foundation.policy import agent_loop
 # Canonical defaults the foundation guarantees. Asserting against named
 # constants keeps the contract explicit (and keeps PLR2004 happy — these
 # are not magic numbers, they are the public API surface).
-EXPECTED_TOOL_CALL_WARN_AT = 50
-EXPECTED_TOOL_CALL_HALT_AT = 150
+EXPECTED_TOOL_CALL_WARN_AT = 100
+EXPECTED_TOOL_CALL_HALT_AT = 300
 EXPECTED_LOOP_THRESHOLD = 3
 EXPECTED_LOOP_WINDOW = 10
 EXPECTED_PM_RESPAWN_MAX_UNPRODUCTIVE = 3