Agent Sandbox is a Kubernetes extension for managing isolated, stateful workloads that run as single long-lived containers with stable identities and persistent storage. It is intended for AI agent runtimes, reinforcement-learning workloads, and other applications that do not fit ordinary stateless deployments. Its catalogue skills and instructions support operating these sandbox workloads.
Borrowing it
Nothing to install: this file belongs to kubernetes-sigs/agent-sandbox. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/kubernetes-sigs/agent-sandbox/main/.agents/skills/fix-flakes/SKILL.mdgit clone --depth 1 https://github.com/kubernetes-sigs/agent-sandboxWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/kubernetes-sigs/agent-sandbox/fix-flakes)<a href="https://agentmods.dev/skills/kubernetes-sigs/agent-sandbox/fix-flakes"><img src="https://agentmods.dev/badge/skills/kubernetes-sigs/agent-sandbox/fix-flakes.svg" alt="Measured on agentmods" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to high
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- high Privilege Escalation · line 39 Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.Fix: Remove references to credential paths. Use environment variables or secrets managers. For docs, use placeholder paths (e.g., /path/to/config). Never load .env or token files in production code paths.
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00080 | $0.01144 |
| Opus 5 | $0.00040 | $0.00572 |
| Sonnet 5 | $0.00016 | $0.00229 |
| Haiku 4.5 | $0.00008 | $0.00114 |
Grade A, and why
fix-flakes scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 8d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 85 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Fix flaky tests from kind/flake issues
Work through open kind/flake issues (filed nightly by dev/tools/flake-report),
reproduce each flake, land a minimal fix, and open a PR that closes the issue.
Evidence before edits: never change a test you could not observe failing or
whose failure output you have not read.
Inputs
- A specific issue number, if the caller gives one; otherwise discover work:
gh issue list --repo kubernetes-sigs/agent-sandbox --label kind/flake --state open --json number,title,body - Skip any issue that already has an open fix PR: search
gh pr list --repo kubernetes-sigs/agent-sandbox --state open --search "<issue number> in:body"and check the results actually referenceFixes #<n>/Closes #<n>. - Skip infra-failure issues (title contains "infrastructure failures") unless
explicitly asked — those are CI tooling work (
dev/ci/, prow job config inkubernetes/test-infra), not test edits. If you do take one, the fix lives indev/ci/shared/runner.pyretries, image prepulls, or job resources.
Reproduce first
- Read the issue: test name, tab/job, failure timestamps, job-history link. Fetch a failing run's junit + build log from the GCS links to read the real error before touching anything.
- Locate the test:
git grep -n "func <TestName>("(Go) orgit grep -n "def <test_name>("(Python SDK e2e). For a Go subtest (TestFoo/subtest), grep for the parent (func TestFoo() only — the part before the first/— then find thet.Run("subtest", ...)inside it. - Stress it:
- Unit tests:
go test ./<package>/... -run '^<TestName>$' -race -count=50(for a subtest,-run '^TestFoo$/^subtest$') - e2e tests need a cluster:
make deploy-kindfirst (see AGENTS.md), thengo test ./test/e2e/... -run '^<TestName>$' -count=10withKUBECONFIG=bin/KUBECONFIG.
- Unit tests:
- If it will not reproduce locally after a reasonable stress run, do NOT guess-fix. Analyze the CI failure output instead; if the cause is still unclear, post your findings as a comment on the issue and stop.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 8d ago First seen · 85 lines · 80 tokens per session scan A b8f1f592dcb3
fix-flakes is a skill published in the GitHub repository kubernetes-sigs/agent-sandbox (3,765 stars, last pushed 3d ago), licensed Apache-2.0. It adds 80 tokens to every session and 1,144 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
fix-failing-tests
Diagnose a failing test in the googleapis/mcp-toolbox repo and land a fix by reasoning from the actual error: read the failure, reproduce it, shrink it until the cause is forced into the open, then fix the cause. Use this whenever a test or CI job is red, a build breaks after a change, many packages fail at once, or a…
troubleshoot-sandbox
Troubleshoot OpenSandbox issues by running diagnostics (logs, inspect, events, summary) via CLI or HTTP API to diagnose sandbox failures like OOM, crash, image pull errors, network problems, etc.
langsmith-observability
LLM observability platform for tracing, evaluation, and monitoring. Use when debugging LLM applications, evaluating model outputs against datasets, monitoring production systems, or building systematic testing pipelines for AI applications.
routing-calibration
Use when calibrating or validating a maintained routing recipe against live model backends, including probe manifests and evidence reports.
verification
Prove that a coding task is actually complete. Use this after meaningful code changes, when tests/builds fail or are skipped, before marking a plan or goal complete, and whenever acceptance depends on runtime, security, recovery, performance, or cross-module evidence.
ulw-qa
This is a Hermes-native ultraqa workflow skill.