Skip to content

Main Navigation Docs CLI Reference

Appearance

Sidebar Navigation

Getting Started

Getting Inference

Choose a Runtime

Configuring GitHub

Per-Org Mode

Repo Management

Operations

Runtimes

Claude Code

Pi

Codex

Agents

Triage

Code

Review

Fix

Retro

Prioritize

Default vs. Custom

Escalation Ladder

User Guides

Adopting Fullsend Incrementally

Bugfix Workflow

Issue Commands

Customizing Agents

Overview

Configuring with AGENTS.md

Configuring with Skills

Configuring Agent Behavior

Bring Your Own Agent

Custom Agent Identity

Config Reference

Harness Field Reference

CEL Triggers Reference

Building custom agents (deprecated)

Running Agents Locally

Jira Integration

How To Emit Traces

Tracing with MLflow

Concepts

Vision

Architecture

Glossary

Reference

Config Reference

Harness Field Reference

Infrastructure

Infrastructure Reference

Mint Administration

Standalone Mint

Private Repositories

Tracing Reference

Eval Measurements

Gate Binaries

Advanced Setup

OpenAI Workload Identity

Layered Config Reference

Contributing

Development

Behaviour Drivers

Behaviour Testing

CLI Internals

E2E Testing

Testing Workflows

Tracing Internals

Contributor Guidelines

Architecture Decision Records (ADRs)

Bot Identities

CEL Triggers and Normalized Events

CI Workflows

Key Design Decisions

Documentation

Forge Abstraction

Go Code

Harness Composition

Harness Field Reference

Mintcore Architecture

Implementing an agent runtime

Sandbox Image Topology

Shell Scripting

Terminology: Tier Conventions

Vouch System

Workflow Contracts

Roadmap

Archived roadmaps

Index

August 2026 roadmap

July 2026 roadmap

June 2026 roadmap

Foundation (April–May 2026)

Landscape

Architecture Decisions

1. Use ADRs for decision making

2. Initial Fullsend Design

3. Org-level configuration lives in a conventional repo

4. Go as the implementation language for core tooling

5. Forge abstraction layer

6. Ordered layer model for install, uninstall, and analyze

7. Per-role GitHub Apps with manifest-based creation

8. workflow_dispatch for cross-repo agent dispatch

9. Use pull_request_target in shim workflows

10. Stored browser session for e2e authentication in CI

11. Canonical schema for admin-managed org config.yaml (v1)

12. Normative v1 Git-tracked files under `.fullsend` for admin install

13. Admin install: repository enrollment v1

14. Admin install: GitHub Apps and `.fullsend` credential surface (v1)

15. Normative implementation contracts live under docs/normative/

16. Unidirectional control flow through the execution stack

17. Credential Isolation for Sandboxed Agents

18. Scripted pipeline for multi-agent orchestration

19. Web source under web/ and Cloudflare project under cloudflare_site/

20. Composable single-responsibility agents with individual sandboxes

21. JSONL reasoning trace exposure from sandboxed agents

22. Harness-level output schema enforcement

23. User documentation structure

24. Harness definitions and shared directory layout

25. Provider-based credential delivery for sandboxed agents

26. Stage-based dispatch for agent workflow decoupling

27. Allowed and Disallowed Tools for Agents

28. GitLab Support Architecture

29. Central token mint and shared apps for a secretless .fullsend

30. OpenShell sandbox interaction model

31. Reusable workflows for action-installed distribution

32. safe-push wrapper binary for sandboxed agents

33. Per-repo installation mode

34. Centralized shim routing via dispatch.yml

35. Layered content resolution

36. Agent Execution Sandbox Architecture

38. Universal harness access via URLs and paths

39. TOTP automation for e2e 2FA

40. Org pool for parallel e2e tests

41. Synchronous workflow_call for event-driven agent dispatch

42. Use /fs- prefix for all slash commands

43. Add upstream source headers to managed scaffold files

44. Deprecate per-org installation mode

45. Forge-portable harness schema

46. Host-side API server design for sandboxed agents

47. Vendored installs with --vendor

48. Automatic Updates

49. Agent configuration environment variable convention

50. Framework-native distributed tracing with OpenTelemetry

47. agent-eval-harness for test infrastructure

48. Functional tests for agent pipelines

53. Agent-driven branch targeting for the code agent

54. Require authorization on all agent dispatch paths

55. Unified environment variable delivery for harness runner and sandbox

56. Per-repo pre-commit tools registry

57. Repos management for per-repo installations

58. Agent registration

59. Public mint mode with wildcard allowlists

60. Cross-org mint authorization via org variables

61. Harness CEL triggers and fullsend dispatch drivers

62. Resolving per-repo dispatch version skew

63. Polling-based work discovery via dispatch drivers

64. Deprecate customized/ directory overlay

65. Provider-backed policy composition

66. Behaviour tests with Gherkin and pluggable drivers

67. GitLab cron-polling event dispatch

68. Public community mint architecture

69. Ready-made configuration presets for simplified installation

70. Portable provider and profile resolution

72. Pre-script skip signalling via the pre-script output protocol

73. Named mint privilege levels under agent roles

74. Repos command consolidation

75. Local-path support for profiles and providers

76. Slash command entity-context separation

77. Mint repos scope hardening with PER_ORG_FOREIGN_COMPAT

78. Simplified mint authorization policy

80. config.yaml vs. agent env vars: where a config option belongs

81. Reserve CI workflow env: for infrastructure plumbing, not agent behavior

82. Separate workflow-host allow-list from caller allow-list

83. Repo-level foreign allow-list

86. Conversation surface for agent participation

87. Eval measurements as online trace scoring with portable export

88. CEL-guarded overlays in the harness schema

89. PR-level risk assessment scoring

90. Runtime-neutral sandbox tool hooks contract

91. Per-agent runtime, model and effort on agents: entries

92. OpenAI WIF credential delivery — runner-side exchange, run-scoped provider

93. Event-source routing for run-status notifications

94. Plugins are runtime-scoped harness resources

95. OpenAI WIF mapping scope — repository-only by default

99. Codex credential delivery — custom model provider with a runner-seeded token file

100. Sandbox tool hooks on Codex — a translating adapter with per-invocation verification

104. Sub-agent model resolution is a runner concern

105. The runner owns the iteration budget

Design Documents

Adaptive Agent Selection

Agent Architecture

Agent-Compatible Code

Agent Infrastructure

Applied Problem Considerations

Architectural Invariants

Autonomy Spectrum

Code Review

Codebase Context

Contribution Volume

Contributor Guidance

Cross-Run Memory

Debugging Agentic Workflows

Downstream/Upstream

Flapping and Convergence

GitLab Support Implementation Details

Governance

Graduated Approval Policy

Human Factors

Intent Representation

MCP Configuration Drift

Operational Observability

Performance and Load Impact Verification

Platform Nativeness

Production Feedback

Repo Readiness

Review Autonomy Evidence

Security Threat Model

Testing the Agents

Tool Call Risk Assessment

Trustworthiness Evidence

Spikes

SPIKE: Community token mint hosting (GCP vs Cloudflare)

Experiments (Exploratory)

1. Agent outage fire drill

2. Claude-based ADR drift scanner

3. ADR-0046 drift scanner

4. Zero-config autonomous bug fix engine with self-improving meta-loop

5. Agent scoped tools triage

6. Code agent evaluation

7. GitHub Actions agent runtime MVP

8. Guardrails evaluation

9. Hermes-inspired security patterns

10. Host-side API server for sandboxed agents

11. Integration Service design doc drift analysis

12. Model Armor vs AI agent triage

13. OpenShell policy bypass

14. OpenShell sandbox evaluation

15. Prompt injection defense-in-depth

16. Promptfoo for agent evaluation in CI

17. Reasoning monitor agent for prompt injection detection

18. Runner hello world

19. Skills

20. Target repository skills in triage

21. Tool scoping

22. Claude GitHub App auth

23. Review cache publication policy

24. CODEOWNERS + GitHub App approval matrix

25. Subagent Process Isolation via PreToolUse Hooks

26. Statistical significance for non-deterministic evals

27. Software Development Life Cycle

Experiment: Agent Evaluation via MLflow + OpenTelemetry

Doc Site

Site Deployment

0.43.0

Other Doc Versions

stable

edge

dev

see all versions

Other Releases

On this page

Getting Started

This section contains easy and to the point guides to help you set up Fullsend. These are intended to be read in a certain order:

Getting Inference
Choose an agent runtime
Configuring GitHub
Organization Mode

Edit this page on GitHub

Next pageGetting Inference