fix(effort): per-role floor for --effort resolution + developer→xhigh

resolve_agent_effort returned '' for all agents in prod because empty
ORCH_AGENT_EFFORT_*= env vars clobber pydantic class-defaults, leaving no
non-empty floor to fall back to -> --effort never reached the Claude CLI.

Add a level-4 per-role floor in resolve_agent_effort (src/agents/launcher.py):
_agent_effort_floor reads the declared class-default of agent_effort_<agent>
(model_fields[...].default), which a present-but-empty env cannot override.
Floor applies only when levels 1-3 are empty and BEFORE validation, so a typo
(non-empty) still drops to '' (never-break ORCH-41) and explicit env/override
still wins (priority preserved). config.py: agent_effort_developer high->xhigh
(single source of truth; floor follows automatically).

Refs: ORCH-081

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
2026-06-08 22:40:39 +03:00
committed by stream
parent 62b4d1f7d1
commit 1ada41f272
5 changed files with 156 additions and 26 deletions

View File

@@ -158,12 +158,50 @@ def resolve_agent_model(agent: str, project_id: str = None) -> str:
return ""
def _agent_effort_floor(agent: str) -> str:
"""ORCH-081 (ORCH-52h): per-role non-empty floor for --effort resolution.
Returns the DECLARED class-default of the ``agent_effort_<agent>`` field on
Settings (e.g. developer -> ``xhigh``, tester/deployer -> ``medium``, the rest
-> ``high``). This is the value pydantic WOULD have used were it not clobbered
by a spurious empty env var (``ORCH_AGENT_EFFORT_<ROLE>=``): the class-default
is fixed in the class body and a present-but-empty env value cannot override it,
so it is a robust floor even when the host ``.env`` zeroes every effort var.
config.py is the single source of truth: upgrading developer to ``xhigh`` there
automatically raises the floor here — no second map to keep in sync (ADR-001).
Unknown agent (a name outside the 6 roles) has no ``agent_effort_<agent>``
field; we degrade to the class-default of ``agent_effort_default`` (``high``),
a safe non-empty floor. Never raises.
"""
fields = type(settings).model_fields
for key in (f"agent_effort_{agent}", "agent_effort_default"):
field = fields.get(key)
if field is not None and field.default:
return field.default
return ""
def resolve_agent_effort(agent: str, project_id: str = None) -> str:
"""ORCH-41: resolve the --effort level for an agent (optionally per-project).
Same priority as resolve_agent_model. The resolved value is validated against
VALID_EFFORTS; an invalid value is logged and dropped (returns "") so a typo
in env/projects_json can never pass a bad flag to the CLI.
Same priority as resolve_agent_model, with one extra level below the global
default (ORCH-081 / ADR-001):
1. project-override (projects_json.agent_efforts[agent])
2. per-agent env (settings.agent_effort_<agent>)
3. global default (settings.agent_effort_default)
4. per-role FLOOR (class-default of agent_effort_<agent>) — NEW
The floor only kicks in when levels 1-3 are all empty (the prod bug: a present
but empty ``ORCH_AGENT_EFFORT_*=`` clobbers every default to ''), guaranteeing
a non-empty target effort for the 6 known roles regardless of host .env state.
The floor is applied BEFORE validation and ONLY to an empty resolve, so it
never masks a typo: an explicit invalid value (e.g. ``turbo``) is non-empty,
skips the floor, and is logged + dropped to "" exactly as in ORCH-41 (the
resolved value is validated against VALID_EFFORTS; an invalid value can never
pass a bad flag to the CLI). Never raises.
"""
value = _resolve_agent_attr(
agent, project_id,
@@ -171,6 +209,11 @@ def resolve_agent_effort(agent: str, project_id: str = None) -> str:
env_attr_prefix="agent_effort_",
default_attr="agent_effort_default",
)
if not value:
# Levels 1-3 all empty (typically a prod .env with empty ORCH_AGENT_EFFORT_*):
# fall through to the per-role floor (class-default). Applied before
# validation but only here, so a typo (non-empty) never reaches this branch.
value = _agent_effort_floor(agent)
if value and value not in VALID_EFFORTS:
logger.warning(
f"Invalid effort '{value}' for agent '{agent}' "

View File

@@ -97,13 +97,15 @@ class Settings(BaseSettings):
agent_model_deployer: str = ""
# ORCH-41: per-agent effort / reasoning level: low|medium|high|xhigh|max.
# Empty -> agent_effort_default. Same resolution order as model. Default split:
# thinking agents (analyst/architect/developer/reviewer) -> high; mechanical
# agents (tester/deployer) -> medium.
# Empty -> agent_effort_default. Same resolution order as model. Default split
# (ORCH-081/ORCH-52h): thinking agents (analyst/architect/reviewer) -> high;
# developer -> xhigh (coding/agentic role, Opus 4.8 canon); mechanical agents
# (tester/deployer) -> medium. These class-defaults are ALSO the per-role floor
# used by resolve_agent_effort when the env is empty (single source of truth).
agent_effort_default: str = "high"
agent_effort_analyst: str = "high"
agent_effort_architect: str = "high"
agent_effort_developer: str = "high"
agent_effort_developer: str = "xhigh"
agent_effort_reviewer: str = "high"
agent_effort_tester: str = "medium"
agent_effort_deployer: str = "medium"