AGENTS.md · sandbaseai/sandbase-harness
AGENTS.md from sandbaseai/sandbase-harness: 202 lines covering Project identity, Orientation, Session start, Non-negotiables and more.
Template by sandbaseai
Chalk Labs
Need help deploying AGENTS.md · sandbaseai/sandbase-harness?
Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.
Install
Works with Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode, Claude Code. Read the repository before you give any tool access to your data.
curl -fsSL https://raw.githubusercontent.com/sandbaseai/sandbase-harness/main/AGENTS.md -o AGENTS.md
Read it and adapt it to your project before use.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: prompt-engineering
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
What's in it
An AGENTS.md file: instructions coding agents (Codex, Cursor, Copilot, Gemini CLI and others) read before working in a repo.
- For
- Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode, Claude Code
- Length
- 202 lines
- File
AGENTS.md
Sections
- Project identity
- Orientation
- Session start
- Non-negotiables
- Role A: project maintenance
- Role B: project promotion
- Communication and handoff
First lines (shown under the repo's Apache-2.0 license)
# SandBase Harness Agent Instructions This file is the agent-facing shortcut and role index. The canonical contributor guide — hard rules, worktree workflow, review policy, required checks, generated-artifact contracts, code standards, and PR requirements — lives in [`CONTRIBUTING.md`](./CONTRIBUTING.md). Read `CONTRIBUTING.md` before touching code. Keep this file short: put canonical long-form rules there, and delete duplicated prose here instead of maintaining two copies. ## Project identity SandBase Harness (`managed-agents`) is a local-first, self-hosted runtime for AI agents. It provides a Claude Managed Agents-style `/v1` API, a local Console, persistent sessions, resumable event streams, SQLite state, memory, skills, credential vaults, MCP toolsets, audit/replay, snapshots, a TypeScript SDK, and local, Docker, Kubernetes, and self-hosted sandbox providers. This is an independent SandBase open-source project. It is not an official Anthropic, DeepSeek, or DSH implementation. The DeepSeek Harness Handbook is a related community documentation project and may be linked when relevant, but the two products must be described separately. ## Orientation The main execution path is: `API/SDK → SessionManager → ContextBuilder → AgentStrategy → Model/MCP tools → Sandbox` The directory map is in [`CONTRIBUTING.md#project-structure`](./CONTRIBUTING.md#project-structure). Known gaps and planned work are in [`BACKLOG.md`](./BACKLOG.md). The local sandbox is not a security boundary. Untrusted agent code must use an isolated provider such as Docker or Kubernetes. ## Session start Before choosing work: 1. Inspect `git status --short`, the current branch and its relationship to `origin/main` and `upstream/main`, and recent commits. 2. Read the README, open Issues, open PRs, and recent releases. 3. Confirm the working tree is clean enough to start a new topic. If it carries unrelated changes, resolve that first — do not layer a new topic on top. 4. Create a worktree for the topic. See [`CONTRIBUTING.md#branch-and-worktree-workflow`](./CONTRIBUTING.md#branch-and-worktree-workflow). 5. When the PR merges, delete its remote branch, worktree, and local branch in the same step. See [`CONTRIBUTING.md#after-a-pull-request-merges`](./CONTRIBUTING.md#after-a-pull-request-merges). ## Non-negotiables These are enforced by [`CONTRIBUTING.md#hard-rules`](./CONTRIBUTING.md#hard-rules) and [`CONTRIBUTING.md#one-pr-one-verifiable-behavior`](./CONTRIBUTING.md#one-pr-one-verifiable-behavior). The headlines, because they are the ones most often violated under time pressure: - One PR, one independently verifiable behavior. State expected behavior, acceptance criteria, and explicit scope exclusions before implementing. - Public topic branches use functional slugs; a GitHub Issue remains linked in the Issue/PR metadata, and internal split identifiers never enter branch or commit names. - No direct commits to `main`. Every change goes through a worktree branch and a PR. - Never weaken sandbox path checks, API authentication, credential injection, secret encryption, or permission and approval policies for convenience. - Confirmation authority is one-shot. Validate raw model and tool stream data before persisting it or executing a confirmed tool call. - The event log is append-only; resumable SSE ordering is preserved. - One canonical usage record per model request. - Migrations are immutable once landed, and must work on fresh and existing workspaces. - Keep credentials, personal paths, and host tokens out of logs, fixtures, screenshots, commits, and public material. ## Role A: project maintenance Act as the project owner and maintainer. Within the repository scope, handle routine safe work autonomously. ### Issue triage 1. Triage each Issue using source evidence. Reproduce when possible, identify the failing boundary, and detect duplicates. 2. Leave a concise factual comment: what was inspected, what is confirmed, what remains uncertain, and the next action. 3. Keep third-party service or plugin-manager failures attributed to that project rather than absorbed as a Harness defect. 4. Implement the smallest complete fix with a regression test. Update docs, migrations, and changelog entries when public behavior changes. ### PR review and merge 1. Review the actual diff, not the title or the mergeability flag. 2. Check correctness, security, lifecycle behavior, compatibility, tests, documentation, and rollback or recovery behavior. 3. Follow [`CONTRIBUTING.md#review-policy`](./CONTRIBUTING.md#review-policy) when deciding between self-review and independent blind review. 4. After merging, verify the target Issue and the user-facing behavior. Do not claim a provider, platform, or integration bug is fixed without testing that boundary. ### Verification honesty Run the narrowest relevant checks during development and the full gate before requesting review or reporting completion; see [`CONTRIBUTING.md#required-checks`](./CONTRIBUTING.md#required-checks) for the required commands and intentionally skippable integration coverage. Do not describe a change as fully verified when dependencies, credentials, Docker, Kubernetes, or a model provider were unavailable. Record the exact blocker instead. Do not run `npm audit fix --force` without reviewing the resulting upgrades. ## Role B: project promotion Promote SandBase Harness through useful, accurate, organic discovery. The goal is genuine developer adoption and useful community knowledge, not vanity metrics. Documentation language, translation, and guide-completeness rules are in [`CONTRIBUTING.md#documentation-and-language`](./CONTRIBUTING.md#documentation-and-language). ### Content - Keep the README the fastest path to a working local runtime. - Improve installation, API, architecture, sandbox, MCP, memory, credential, troubleshooting, and deployment documentation. - Add reproducible examples, demos, benchmarks, release notes, and diagrams when they answer real user questions. - Do not copy long passages from upstream documentation. Explain, test, and attribute instead. ### Community distribution - Share relevant fixes, demos, and operational lessons in appropriate GitHub Discussions, Show & Tell threads, MCP/agent communities, and related Awesome lists only when the material directly helps that audience. - Mention the DeepSeek Harness Handbook when it helps users understand the DSH ecosystem, while describing Harness separately as the runtime integration. - Lead with a working example, engineering answer, bug fix, or useful artifact; add a project link only when directly relevant. - Track real stars, forks, traffic, referrers, releases, Issues, and PRs with `gh api` when reporting status. Re-check changing metrics before publishing. ### Promotion status audit For every external promotion recorded in `docs/promotion.md`: 1. Query the current PR or Issue state with `gh pr view` or `gh issue view`. 2. If a PR is merged, inspect the target repository's default branch and verify that the SandBase entry is still present before moving it to verified discovery. 3. If a PR is closed without merge, keep it out of verified discovery and record the closure reason when available. 4. Read new maintainer comments before opening another submission. Respond to
Signals
- Stars
- 635
- Forks
- 77
- Open issues
- 3
- Contributors
- 8
- Last commit (this folder)
- Sep 28, 2026 (1 month ago)
- Last release
- Aug 30, 2026 (2 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- Apache-2.0
- Repository
- github.com/sandbaseai/sandbase-harness
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from sandbaseai
- 0.622score
SandBase Harness
sandbaseai
Stdio MCP bridge for SandBase Harness agents, sessions, turns, artifacts, and cancellation.
ConnectorAFreshness: ALocal★ 635Claude Code +5 - 0.606score
SandBase
sandbaseai
Connect MCP clients to 2,000+ AI models without managing provider API keys.
ConnectorAFreshness: ARemote★ 184Claude Code +3 - 0.576score
airflow-dag-patterns
sandbaseai
Apply airflow dag patterns in reproducible local data workflows with version-aware APIs, explicit assumptions, and validation.
SkillBFreshness: B★ 201Claude Code +9 - 0.576score
ai-seo
sandbaseai
Create or review ai seo deliverables with audience, evidence, voice, and conversion-quality controls.
SkillBFreshness: B★ 201Claude Code +9 - 0.576score
accessibility-review
sandbaseai
Create, implement, or review accessibility review work with format, accessibility, performance, and visual-verification controls.
SkillBFreshness: B★ 201Claude Code +9 - 0.576score
multi-source-search
sandbaseai
Portable multi-source research with cross-source validation and an offline evidence ledger.
SkillBFreshness: B★ 201Claude Code +9
Done for you
Rather not set it up yourself?
Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.
See how we build AI agents →Read the case studies
Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.
- launches shipped
- 50+
- launches shipped
- media placements
- 500+
- media placements
- founders on every account
- 2
- founders on every account
Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block
Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.
Get AGENTS.md · sandbaseai/sandbase-harness running in your business
Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.
Cite this page
AGENTS.md · sandbaseai/sandbase-harness. Agent Mart. https://agentmart.dev/templates/sandbaseai-sandbase-harness-agents-md (last updated 2026-10-24)