drift
AGENTS.md14 starsChanged 7 months agoMITView on GitHub
This file is the agent-neutral entry map for the Drift repository. It is a map, not a rulebook: follow the linked source files for the operative rules. The harness is distributed across MCP tools, session state, benchmarks, and agent-facing docs: Machine-near harness rules live in docs/agent-harness-golden-principles.md. Add or tighten a mechanical check whenever a rule can be verified by code, tests, or CI. The current harness audit deliverables are versioned under audit/: Use these commands for local agent-harness changes: For…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,700 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository41 other files it gives its agents.
- .github/AGENTS.mdAGENTS.md
- .github/copilot-instructions.mdCopilot instructions
- llms.txtllms.txt
- brainstormingSkill
- drift-adr-workflowSkill
- drift-agent-prompt-authoringSkill
- drift-commit-pushSkill
- drift-dependency-updateSkill
- drift-docs-gap-fixSkill
- drift-evidence-artifact-authoringSkill
- drift-finding-message-authoringSkill
- drift-ground-truth-fixture-developmentSkill
- drift-mcp-repo-ops-usageSkill
- drift-pr-review-loopSkill
- drift-pr-reviewSkill
- drift-releaseSkill
- drift-risk-audit-artifact-updatesSkill
- drift-security-triageSkill
- drift-session-handover-authoringSkill
- drift-signal-development-full-lifecycleSkill
- fix-failing-actionsSkill
- grill-meSkill
- guard-src-drift-apiSkill
- guard-src-drift-commandsSkill
- guard-src-drift-ingestionSkill
- guard-src-drift-outputSkill
- guard-src-drift-signalsSkill
- guard-src-driftSkill
- improve-codebase-architectureSkill
- prd-to-issuesSkill
- self-improving-agentSkill
- simplify-and-hardenSkill
- skill-creatorSkill
- subagent-driven-developmentSkill
- systematic-debuggingSkill
- tddSkill
- test-driven-developmentSkill
- verification-before-completionSkill
- workspace-automation-orchestratorSkill
- write-a-prdSkill
- writing-plansSkill
The file
We split this AGENTS.md into the rules it gives an agent
5 rules. About 0 are common, 4 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Read the mandatory agent rules in .github/copilot-instructions.md.
- Very specific: line 9
Run the policy gate from .github/instructions/drift-policy.instructions.md before file work.
- Distinctive: line 10
Use DEVELOPER.md for setup, commands, architecture overview, and targeted tests.
- Distinctive: line 11
Use .github/AGENTS.md for the prompt, skill, and evaluation-prompt catalogue.
- Distinctive: line 42
Use these commands for local agent-harness changes:
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Drift Agent Root Map
This file is the agent-neutral entry map for the Drift repository. It is a map, not a rulebook: follow the linked source files for the operative rules.
Start Here
- Read the mandatory agent rules in .github/copilot-instructions.md.
- Run the policy gate from .github/instructions/drift-policy.instructions.md before file work.
- Use DEVELOPER.md for setup, commands, architecture overview, and targeted tests.
- Use .github/AGENTS.md for the prompt, skill, and evaluation-prompt catalogue.
Harness Engine Map
The harness is distributed across MCP tools, session state, benchmarks, and agent-facing docs:
| Area | Primary artifacts | Contract |
| MCP entrypoint | src/drift/mcp_server.py, .vscode/mcp.json | Register tools and delegate business logic to router modules. |
| Session loop | src/drift/session.py, src/drift/mcp_orchestration.py, src/drift/mcp_router_session.py | Preserve state, trace, queue replay, guardrails, and handover gates across turns. |
| Agent tasks | src/drift/models/_agent.py, src/drift/output/agent_tasks.py | Emit machine-readable repair tasks with constraints and verification plans. |
| Evaluation harness | scripts/ab_harness.py, scripts/benchmark_agent_loop.py, scripts/run_agent_loop_benchmark.py | Measure agent loop efficiency, severity-gate routing, and brief effectiveness. |
| Mechanical harness check | scripts/check_agent_harness_contract.py, tests/test_agent_harness_contract.py | Verify required docs, audit artifacts, root allowlist entries, Markdown links, and MCP boundaries. |
Golden Principles
Machine-near harness rules live in docs/agent-harness-golden-principles.md. Add or tighten a mechanical check whenever a rule can be verified by code, tests, or CI.
Audit Trail
The current harness audit deliverables are versioned under audit/:
- audit/harness-engine-audit.md - current-state map and risks
- audit/improvement-plan.md - prioritized backlog and validation method
- audit/change-log.md - implemented changes and checks
- audit/follow-up.md - gaps that still need design or automation
- audit/harness-tool-map.json - canonical MCP tool to router-owner map (HARNESS006/007)
Validation Commands
Use these commands for local agent-harness changes:
.venv\Scripts\python.exe scripts\check_agent_harness_contract.py --root .
.venv\Scripts\python.exe -m pytest tests/test_agent_harness_contract.py -q --tb=short
make agent-harness-checkFor source-file-specific checks, use the map in DEVELOPER.md or run:
make test-for FILE=src/drift/session.pyDoes it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.