forkless/strix
1
0
Fork 0
mirror of https://github.com/usestrix/strix.git synced 2026-08-18 17:52:32 +02:00
Code Issues Packages Projects Releases Wiki Activity

feat(runtime): graduated wrap-up warnings, budget reserve, and interactive budget pause/continue (#893)

Browse Source 
Co-authored-by: Ahmed Allam <ahmed39652003@gmail.com>
This commit is contained in:
devin-ai-integration[bot]
2026-07-26 20:37:14 -07:00
committed by GitHub
co-authored by Ahmed Allam
parent 47617969d3
commit c55a8fa4ba
12 changed files with 1628 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/usage/cli.mdx
+36 -3
View File
@@ -61,11 +61,28 @@ strix (--target <target> | --target-list <path> | --mount <path>) [options]
Path to a custom config file (JSON) to use instead of `~/.strix/cli-config.json`.
</ParamField>
<ParamField path="--max-budget-usd" type="number">
<ParamField path="--max-budget" type="number">
Maximum LLM spend in USD for the whole scan, counted cumulatively across the
root agent and every child agent. The budget is checked after each model
response; once the running cost reaches the threshold, the scan stops cleanly
with a `stopped` status (not a failure) and the sandbox is torn down.
response.
In non-interactive mode (`-n`), once the running cost reaches the threshold,
the scan stops cleanly with a `stopped` status (not a failure) and the sandbox
is torn down. Sub-agents are stopped early, at 90% of the budget, reserving
the final slice for the root agent to wind down and produce the final report.
In interactive mode, reaching the budget pauses the scan instead of ending
it: every agent parks, and sending any message resumes the scan with the cap
extended by the original budget amount. There is no sub-agent reserve in
interactive mode.
As the budget is approached, graduated wrap-up warnings are surfaced to
**every** agent so they can finish their work and call their lifecycle tool
before the hard stop. The bands sit just below each role's own stop point: the
root is warned at **70%, 85% and 95%** (it stops at 100%), while sub-agents are
warned at **75%, 80% and 85%** (they stop at the 90% reserve). In interactive
mode every agent uses the **70%, 85% and 95%** bands. Percentages shown in the
warnings are the real cumulative spend against the full budget.
Must be greater than `0`. Omit the flag for no limit.
@@ -84,6 +101,19 @@ strix (--target <target> | --target-list <path> | --mount <path>) [options]
counts.
</ParamField>
<ParamField path="--max-turns" type="integer" default="500">
Maximum number of turns (one model response plus its tool round) allotted to
**each** agent, applied per run. When an agent reaches this limit it is
force-stopped.
As the limit is approached, graduated wrap-up warnings (at 70%, 85% and 95%)
are injected into that agent's next model turn so it can prioritise its
remaining work and call its lifecycle tool (`finish_scan` for the root agent,
`agent_finish` for sub-agents) before the hard stop.
Must be greater than `0`.
</ParamField>
## Examples
```bash
@@ -99,6 +129,9 @@ strix --target api.example.com --instruction "Focus on IDOR and auth bypass"
# CI/CD mode
strix -n --target ./ --scan-mode quick
# Cap cost and per-agent turns
strix --target https://example.com --max-budget 25 --max-turns 300
# Force diff-scope against a specific base ref
strix -n --target ./ --scan-mode quick --scope-mode diff --diff-base origin/main