| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
| |
This seems to help a bit with number of round-trips.
|
| |
|
|
|
|
| |
Calculating total disk usage of all cache entries can take very long time
for large caches (needs to stat all files). This is especially problematic
for tools/syz-aflow. Cache disk usage in the meta file.
|
| |
|
|
|
| |
We currently duplicate list of source extensions in the build action
and codesearch tool. Unify the lists.
|
| |
|
|
|
|
|
|
|
|
| |
The error allows tools to communicate that an error is not an infrastructure error
that must fail the whole workflow, but rather a bad tool invocation by an LLM
(e.g. asking for a non-existent file contents).
Previously in the codesearcher tool we used a separate Missing bool
to communicate that. With the error everything just becomes cleaner and nicer.
The errors also allows all other tools to communicate any errors to the LLM
when the normal results cannot be provided and don't make sense.
|
| |
|
|
| |
Just provides full file contents as last resort.
|
| |
|
|
|
| |
dir-index provides a list of subdirectories and files in the given
directory in the source tree.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
Gracefully handle (reply to LLM with error):
- incorrect tool name
- incorrect tool arg type
- missing tool arg
Silently handle:
- more than one call to set-results
- excessive tool args
Fixes #6604
|
| |
|
|
|
|
|
|
| |
Detect model quota violations (assumed to be RPD).
Make syz-agent not request jobs that use the model
until the next quota reset time.
Fixes #6573
|
| |
|
|
|
|
| |
Having LLM model per-agent is even more flexible than per-flow.
We can have some more complex tasks during patch generation with the most elaborate model,
but also some simpler ones with less elaborate models.
|
| |
|
|
| |
Update #6573
|
| |
|
|
|
|
|
|
|
| |
Add LLMAgent.Candidates parameter.
If set to a value N>1, then the agent is invoked N times,
and all outputs become slices.
The results can be later aggregated by another agent,
as shown in the test.
|
| |
|
|
|
|
|
|
|
|
| |
Flow errors denote failure of the flow itself,
rather than an infrastructure error. A flow errors mean an expected
condition in the flow when it cannot continue, and cannot produce
expected outputs. For example, if we are doing something with the kernel,
but the kernel build fails. Flow errors shouldn't be flagged in
Fixes #6610
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
CitationMetadata may be present in replies sometimes.
CitationMetadata is a specific field in the Gemini API's response object
that alerts you when the model has directly quoted or closely derived
content from a specific source, such as a book, website, or open-source code repository.
We've got the following error:
syz-agent: unexpected reply fields ({Content:0xc0058eb4a0 CitationMetadata:0xc0094009a8
FinishMessage: TokenCount:0 FinishReason:STOP AvgLogprobs:0
GroundingMetadata:<nil> Index:0 LogprobsResult:<nil> SafetyRatings:[]
URLContextMetadata:<nil>})
|
| |
|
|
|
| |
We may want to use a weaker model for some workflows.
Allow to use different models for different workflows.
|
| |
|
|
|
| |
Add race:harmful/benign label.
Set it automatically by confirmed AI jobs.
|
| |
|
|
|
|
| |
Add workflow that can be used for moderation of UAF bugs (consistent/actionable reports),
such UAF bugs can be upstreammed automatically, even if they happened only once
and don't have a reproducer.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Rephrase the prompt to be only about KCSAN,
currently it has some leftovers from more generic assessment prompt
that covered KASAN bugs as well (actionability).
Also add Confident bool output.
We may want to act on both benign/non-benign,
so we need to know when LLM wasn't actually sure either way.
This should also be useful for manual verification/statistics.
If LLM is not confident and can can admit that, it's much better
than giving a wrong answer. But we will likely want to track
percent of non-confident answers.
|
| |
|
|
|
| |
Paths passed to filepath.WalkDir are absolute (include the dir prefix), account for that.
Strings returned by filepath.Ext include the dot, account for that as well.
|
| |
|
|
|
| |
Keep vmlinux for debug info/addr2line.
Keep compile_commands.json, we need it for codesearch.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
