fix: keep -p stdout as the answer, not turn-pulse chrome - #657
Closed
justrach wants to merge 1 commit into
Closed
Conversation
Rematch schema-output wrote the right object but json.load failed because `· turn still going ·` rode g_out. emitNotice already dropped --json; -p sets unattended and promised stdout was only the answer. Also tell the agentic --output-schema turn not to hunt .graff for a schema file (ADR 0001 still keeps the grammar off that turn).
justrach
marked this pull request as ready for review
August 28, 2026 05:35
Owner
Author
|
Already on main (tool_pulse keeps -p stdout as the answer). Closing as superseded. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Rematch trajectories (280 + catalog rebuild, SuperGrok, grok-4.6) said the next win was not another catalog bug.
What the traces showed
Shared tasks stayed 21/22 vs 21/22. Graff is still faster and cheaper. The leftover miss and the leftover waste were both on stdout / “schema” wording, not on tools.
-pstdout· turn still going ·.schema-outputwrote{"distinct_skus":4,"top_sku":"SKU-B"}thenjson.loadfailed because four pulse lines sat in front.runOneshotPromptalready promised “stdout carries only the answer”;emitNoticestill wrote tog_out..graff/codedb list_dir+head inventory.txt(enough), thenfind ./head .graff/sessions/…looking for a schema file. Two extra bash/API rounds. Prefix already mentions.graff/traces.exact-reply(“pong”) is 8013 in. Stable across the 22 tasks. Catalog + system prompt, not the fixture.projectRootIdis cwd-derived. Sibling sandboxes never share the first-call cache. Grok’s first call often arrives already warm (11k–43k cached).label-sortboth miss_natural_first("+2","2"). Same fail on grok.cookie-storeSIGABRT under-j 6This PR
tool_pulse.emitNoticedrops chrome whenunattended(-p) the same way it already drops--json. Hosted TUI/ACP still get the pulse viahostedSink.--output-schemaoneshots get one agentic-phase line: a later step applies the schema; do not search for a file and do not open.graff/. The formatting turn is unchanged.Not in this PR
Slimming the 8k prefix, sharing a cache key across eval cwds, or the cookie-store abort. Those need their own A/B.