Skip to content

eval(harbor): trim removed-behavior history from the settings docstring #752

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ericleepi314 merged 1 commit into from
Jul 26, 2026
+3 −9
Merged

eval(harbor): trim removed-behavior history from the settings docstring #752

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 3 additions & 9 deletions eval/harbor/clawcodex_agent.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -456,15 +456,9 @@ async def _seed_container_settings(
— the global-config path deliberately does not follow
CLAWCODEX_CONFIG_DIR) makes the requested effort session-wide.

This seeds NOTHING but effort. An earlier version also appended
Harbor's wall-clock deadline to the system prompt ("stop broad
exploration, switch to the narrowest checks"). That is harness-shaped
guidance no other agent receives: the built-in claude-code adapter
appends nothing, so it made clawcodex-vs-Claude-Code trajectories
non-comparable, and it would not exist on a submitted run scored by
the official tooling. Any behavior worth having under a deadline
belongs in the product's own prompt, applied to every task, not
injected by an eval adapter.
Effort is the ONLY thing seeded here: guidance an eval adapter gives
one agent and not another does not belong in a comparison, and would
not exist on a run scored by the official tooling.
"""
settings: dict[str, Any] = {}
effort = self._resolved_flags.get("effort")
Expand Down
Loading