Commit graph

873 commits

Author SHA1 Message Date
Jonathan Singer
098fa3659c fix: guard resumed assessments and ignore withdrawn fix records 2026-10-01 14:45:24 -04:00
Jonathan Singer
f14468565a fix: complete interactive autofix scans before cleanup 2026-10-01 14:23:56 -04:00
yoni
1fa7d211b4 feat: publish verified fixes as local branches 2026-10-01 15:16:20 +00:00
yoni
3763a677f9 fix: finalize cancelled fix agents 2026-10-01 13:42:30 +00:00
yoni
60d4ce15e5 feat: allow scans to skip automatic fixes 2026-10-01 07:07:06 +00:00
Jonathan Singer
bba4aa24dc fix: require reviewed current patches and enforce fix network isolation 2026-09-30 23:21:24 -04:00
yoni
f75fb5fcf9 fix: harden fix dispatch and verification (STR-815) 2026-10-01 02:34:07 +00:00
Jonathan Singer
1f8295c681 Launch native fixes after persistence and bound completion failures 2026-09-30 19:04:22 -04:00
Jonathan Singer
b71ed13f4d Use native child delegation for finding fixes and strengthen completion guidance 2026-09-30 17:07:08 -04:00
Jonathan Singer
83176652f5 Run confirmed finding fixes as native agents in the scan sandbox 2026-09-30 12:55:01 -04:00
yoni
17894006b4 Keep an approved fix when only the PR text changes 2026-09-30 08:24:08 +00:00
yoni
ba6bbaf9c6 Include repair follow-ups in the readable review 2026-09-30 08:21:23 +00:00
yoni
e4f1fe6cfc Merge remote-tracking branch 'origin/main' into devin/1790308365-verified-fix-preparation 2026-09-30 08:18:59 +00:00
Jonathan Singer
868ba53927 Scope fix validation and warn on repeated commands 2026-09-30 03:47:36 -04:00
Jonathan Singer
77bd5dade5 Require an explicit fix handoff for source-backed findings 2026-09-30 03:40:21 -04:00
Ahmed Allam
ef272b8e0d chore(models): remove the model quality warning and its allowlists 2026-09-30 10:04:28 +03:00
Jonathan Singer
1c1a899b3c Focus fix agents and preserve completion evidence 2026-09-30 01:28:56 -04:00
Jonathan Singer
acf262f1c6 Keep fix outputs private and outside the source checkout 2026-09-30 01:07:43 -04:00
Jonathan Singer
d35197bea1 Move complete fix workflow into OSS and add strix fix CLI 2026-09-30 00:43:53 -04:00
Jonathan Singer
5badb2d541 Remove retired fix execution and proof machinery 2026-09-29 23:26:23 -04:00
Jonathan Singer
348fbf2608 Support native fix-agent assignments and final reviewed patches 2026-09-29 22:36:47 -04:00
Ian
9b72488c92 docs + unit test fix 2026-09-30 04:46:01 +03:00
Ian
c77bdd2c2a Disable by default 2026-09-30 04:46:01 +03:00
Ian
c814f6bf30 Made session IDs optional 2026-09-30 04:46:01 +03:00
Ian
a1658fe59b Scarf support 2026-09-30 04:46:01 +03:00
Ian
3fbccc6b1d Non-streaming path 2026-09-30 04:46:01 +03:00
Ian
c9aebc6c87 larger default block size 2026-09-30 04:46:01 +03:00
Ian
95fbd8d687 Openrouter sticky sessions for caching, with telemetry 2026-09-30 04:46:01 +03:00
Ahmed Allam
463b149bdb fix(budget): parked agents count as active; park never overwrites a stop
active_agents_except (finish_scan, wait_for_message) treats budget_paused as
active, so a root cannot finish the scan over a parked child. park_for_budget
only transitions a running agent, and the wake back to running happens under
the coordinator lock.
2026-09-30 03:13:40 +03:00
Ahmed Allam
d355838ea0 feat(budget): budget_policy=pause parks every agent at the limit until the operator resumes
Adds budget_policy: stop | pause to run_strix_scan / ReportUsageHooks /
AgentCoordinator, independent of interactive mode. Under pause the agents
get no budget warnings and no sub-agent reserve; each agent parks before
its next LLM call once spent >= limit or the scan is paused, sessions and
sandbox stay alive, and coordinator.resume_budget(max_budget_usd=...)
replaces the limit and wakes every parked agent without adding anything
to any session. coordinator.pause_budget() parks a running scan the same
way. In-flight calls are never cancelled, so spent may end above the
limit. Parked agents count as active for stop_agent.
2026-09-30 03:13:40 +03:00
Jonathan Singer
d184142377 Let fix reviewer own validation and completion 2026-09-29 17:59:58 -04:00
ian-at-strix
0ff9f8c324
feat(tui): animate the wait_for_agents indicator (#1383) 2026-09-29 14:39:07 -07:00
ian-at-strix
954bc0d527
perf(prompt): load requested skills after a cache point (#1382)
Siblings differ only in the skills they were spawned with, but those came
first in <specialized_knowledge>, so their prompts diverged at 39%. Shared
skills and the catalog now come first, and the requested skills follow a
cache point, so siblings share 93%.

The extra system message takes a fourth Claude breakpoint, so the Bedrock
tool_config one goes: the first system breakpoint already covers the tools.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-30 00:35:22 +03:00
ian-at-strix
e66c56c473
perf(llm): give Claude a cache point before the per-run scope (#1376)
* perf(llm): give Claude a cache point before the per-run scope

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(llm): split the system prompt at a generic <cache_point> marker

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-30 00:35:22 +03:00
ian-at-strix
50425c2c99
perf(prompt): put per-run scope at the end of the system prompt (#1375)
* perf(prompt): put per-run scope at the end of the system prompt

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* test(prompt): assert scope renders once, after the shared prefix

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-30 00:35:21 +03:00
Jonathan Singer
43391ebefa Let repair and review agents own the fix workflow 2026-09-29 16:22:49 -04:00
Jonathan Singer
0faa7b7da1 Make fix handoffs actionable and require customer unit tests 2026-09-29 13:27:27 -04:00
alex s
6ae036e6c1
docs(skills): refresh framework behavior and security testing guidance (#1372) 2026-09-29 10:11:02 -07:00
Jonathan Singer
f03fd72d79 Simplify fix preparation around native tests and independent review 2026-09-29 11:49:00 -04:00
Jonathan Singer
85b3030816 Preserve partial fixes and require consistent execution evidence 2026-09-29 05:31:58 -04:00
yoni
791ef9106d fix: let independent evidence resolve repair timeout 2026-09-29 08:22:18 +00:00
Jonathan Singer
198a254be8 Make preparation history factory explicit for strict type checking 2026-09-29 03:56:52 -04:00
Jonathan Singer
f213a7da6a Require functional fix evidence and preserve partial preparation work 2026-09-29 03:53:31 -04:00
ian-at-strix
0c702723aa
fix(tui): suspend on ctrl+z (#1371)
Bubble Tea's raw mode clears ISIG, so ctrl+z reached the TUI as a key and
was ignored instead of stopping the job. Return tea.Suspend for it on every
screen, and re-enable mouse tracking on resume, since Bubble Tea's restore
brings back the alt screen but not the mouse mode.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-28 22:44:25 -04:00
yoni
00fcb4c93a fix: classify distinct check failures correctly 2026-09-28 12:41:23 +00:00
yoni
df313c1438 refactor fix preparation gates 2026-09-28 12:13:19 +00:00
yoni
c510f58ef3 fix: retry transient verifier inconclusive results 2026-09-28 01:53:04 +00:00
Yash Aslekar
d6dd9dec26
fix(dev): make check-all non-mutating (#1360) 2026-09-27 17:47:20 -07:00
yoni
a41000d5bf Retry unchanged repairs when verification changes 2026-09-25 21:57:34 +00:00
yoni
d02b74c3c0 Preserve explicit blocked repair outcomes 2026-09-25 21:12:49 +00:00