Tracer-Cloud/opensre

Context Follow-up to #5844 (macOS/Linux curl install). On Unix, opensre after a fresh install via the cURL script can take 7–14 seconds to load. We need the same cold-start measurement on Windows so we know whether the slowdown is installer-specific, platform-specific, or shared (e.g. Python import / binary startup). Requested in Discord — Windows counterpart to #5844. This issue is investigation only — collect timings and a short root-cause hypothesis. Fix work can be a separate follow-up once we have Windows numbers. Goal Measure how long opensre takes to become usable after a clean Windows install, and identify where time is spent. Replicate (Windows) Remove OpenSRE from the machine (binary/shims on PATH, any venv, and local checkout if present). Fresh install via the README PowerShell one-liner: irm https://install.opensre.com | iex Open a new PowerShell window (so PATH is current). Time the first opensre invocation, e.g.: Measure-Command { opensre --help } Also note

Problem Two engineer-facing docs describe the session-goal loop as the "outer loop": File Line core/agent_harness/AGENTS.md "uses dispatch, not the SessionGoal outer loop" gateway/AGENTS.md "agent.handle(text, TurnBinding(...)) — SessionGoal outer loop + ..." "Outer" only means something once you know there is an inner one, and the doc never says what the inner one is. Why it matters There are two real loops with real names: the SessionGoal turn budget (how many turns a goal may span) and the ReAct loop inside a single turn (tool calls until the model stops). "Outer" and "inner" are the informal shorthand for those, and a reader who has not sat in the conversation where the shorthand was coined cannot decode it. Both sentences already say "SessionGoal" — the word "outer" adds nothing. Steps In each sentence, replace "outer loop" with the thing it actually names — the SessionGoal turn loop, or simply "SessionGoal turns". Where the sentence contrasts it with what happe

Problem core/agent_harness/session_goal/run_until.py has two private helpers named after metaphors rather than what they do: Now Should say _paint(...) _emit_progress (it attaches the goal and calls on_progress) _finish_outer_turn(...) _finish_session_goal_turn Both are module-private with no importer outside this file — six calls to _paint, three to _finish_outer_turn. Why it matters "Outer turn" only means something once you know there is an inner one. The package's own vocabulary is SessionGoal turn (the budget) versus ReAct goal (the loop inside a turn), and the private names are the last place still using the older insider phrasing. Because both helpers are private and local, this is one of the safest renames in the tree — no public API, no persisted value, no wire format. Steps Rename _paint to a progress verb and update its six call sites. Rename _finish_outer_turn → _finish_session_goal_turn and update its three call sites. Rename the local variable painted i

Problem core/agent_harness/session_goal/continuation.py exports continuation_nudge. A "nudge" is chat jargon; the function returns the user-message text that continues a session goal, which is a prompt. Ten references across the tree, all first-party. Why it matters .cursor/rules/no-jargon-names.mdc and docs/NAMING.md ask for names that state behaviour. continuation_prompt tells a reader what comes back (prompt text); continuation_nudge tells them how the author felt about it. Steps Rename the function continuation_nudge → continuation_prompt in core/agent_harness/session_goal/continuation.py. Update __all__ in that module and any re-export in core/agent_harness/session_goal/__init__.py. Update the call site in run_until.py (the import near the top and the call near the bottom). Update tests. Fix the prose in the files you touched — a docstring that still says "nudge" next to a function named continuation_prompt is worse than before you started. Only in files this rename touches.

Problem Three docstrings and one contributor tip describe a package's public API as a "door". .cursor/rules/no-jargon-names.mdc and docs/NAMING.md ask for behaviour nouns a stranger can read; "door" is a metaphor that has to be learned before it means anything. File Line core/tool_framework/__init__.py "The tier's door for consumers above it." core/tool_framework/__init__.py "...behind the sibling core.tool_framework.utils door" core/tool_framework/utils/__init__.py "the one door consumers import these helpers through" AGENTS.md (adding a tool, step 1) "Import through those tier doors, not their internal submodules" Why it matters This is the first thing a contributor reads when adding a tool, and the sentence it appears in is otherwise precise — it names the border test that enforces the rule. "Door" is the one word in it that carries no meaning on first read. Steps Replace each occurrence with a plain term: public API, package entry, or import facade. Pick per sent