Skip to content

Record D-5, D-6 and D-7 (#56, #57, #58) #59

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鈥檒l occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
shipmill-romamo merged 1 commit into from
Oct 9, 2026
Merged

Record D-5, D-6 and D-7 (#56, #57, #58) #59

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
24 changes: 24 additions & 0 deletions docs/decisions.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,3 +35,27 @@ entry that supersedes it, never by editing an old one.
- Why: The model's reasoning quotes host output it read, so unbounded reasoning makes the record unbounded and stores host output in it
- Applies to: sdk/src/cloudfall/investigation.py, config/schemas/v1/agent-investigation.schema.json
- Enforced by: review

## D-5: agent investigate exits PRECONDITION when its record can't be written

- Decided: 2026-10-09, in cloudfall-dev/cloudfall#56
- Rule: When agent investigate cannot write the investigation record, it exits PRECONDITION with no traceback and an error naming the investigations folder; this holds for a stopped run too, whose record is then lost but whose exit stays non-zero
- Why: An unwritable store folder is the same kind of refusal as a bad --decisions folder, and a traceback with exit 0 breaks D-3
- Applies to: sdk/src/cloudfall/investigation.py, sdk/src/cloudfall/app.py, agent investigate
- Enforced by: review

## D-6: The investigation answer and raw tool-call arguments are capped from the start

- Decided: 2026-10-09, in cloudfall-dev/cloudfall#57
- Rule: The investigation record keeps at most 8000 characters of spec.answer and of each trace[].toolCalls[].arguments, keeping the start and marking the cut; the schema bounds both fields
- Why: Both can quote host output like the reasoning D-4 caps, and the finding sits at the start of the answer, so the start is what to keep
- Applies to: sdk/src/cloudfall/investigation.py, config/schemas/v1/agent-investigation.schema.json
- Enforced by: review

## D-7: Replayed and live proposals never supersede each other

- Decided: 2026-10-09, in cloudfall-dev/cloudfall#58
- Rule: A newer proposal supersedes an older proposed record (D-1) only when both have the same replay value, so a replay never closes a live proposal and a live run never closes a replayed one; a replayed read keeps status ran and its verdict says the output was played back, with no schema change
- Why: A replayed record can't be approved (D-2), so letting it supersede a live proposal leaves the operator with nothing to approve
- Applies to: sdk/src/cloudfall/decision.py, sdk/src/cloudfall/app.py, agent investigate --replay, cloudfall why
- Enforced by: review
Loading