next-level-agent
AGENTS.md7 starsChanged 5 months agoMITView on GitHub
This repository contains Next Level Agent for OpenCode. These instructions are for a coding agent, including OpenCode or Codex, that is helping a user install, verify, update, or develop NLA. Codex may assist with the installation process, but the complete NLA runtime is currently supported only in OpenCode. Do not describe this file as evidence that NLA supports Codex CLI or any other coding-agent CLI. Install NLA as a stable local clone and launch OpenCode against the user's selected project…
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 31,774 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 repository19 other files it gives its agents.
- compactSkill
- assistant-notebookSkill
- brainstormingSkill
- dispatching-parallel-agentsSkill
- executing-plansSkill
- finishing-a-development-branchSkill
- mem0-memorySkill
- next-level-agentSkill
- planning-parallel-workSkill
- receiving-code-reviewSkill
- requesting-code-reviewSkill
- subagent-driven-developmentSkill
- systematic-debuggingSkill
- test-driven-developmentSkill
- using-git-worktreesSkill
- verification-before-completionSkill
- writing-acceptance-testsSkill
- writing-plansSkill
- writing-skillsSkill
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
25 rules. About 2 are common, 18 are distinctive, and 5 are very specific to this project.
≈2
Common · estimate
Many files like this have a rule of this kind.
≈18
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
Read README.md for product architecture and role boundaries.
- Distinctive: line 14
Read INSTALL.md for the supported Alpha installation procedure.
- Very specific: line 15
Read docs/PROJECT_STATUS_AND_USAGE.md for current limitations, telemetry, state locations, and roadmap.
- Very specific: line 16
Read docs/DRAFT_0_4_IMPLEMENTATION_STATUS.md before proposing work from the historical Draft 0.4.
- Distinctive: line 32
Ask which project the user wants to run NLA against if it is not clear from the current workspace.
- Very specific: line 33
Use a stable clone path such as ~/.local/share/nla/next-level-agent.
- Common: line 34
Do not overwrite an existing clone, global OpenCode config, project config, shell profile, provider config, or credentials.
- Distinctive: line 35
Before updating an existing clone, run git status --short --branch and preserve local changes.
- Distinctive: line 36
Do not run destructive Git commands.
- Common: line 37
Do not store API keys in the repository or NLA state.
- Distinctive: line 38
Do not change model bindings without showing the proposed change to the user when it affects provider access or cost.
- Distinctive: line 39
Do not install packages or providers unless the user authorizes it.
- Very specific: line 40
Prefer OPENCODE_CONFIG over copying the complete NLA configuration into every project.
- Very specific: line 41
Treat opencode debug config as a required non-LLM verification step.
- Distinctive: line 42
Use a harmless read-only smoke test before a real task.
- Distinctive: line 62
Check that every enabled pool has at least one model the user can access.
- Distinctive: line 73
Verify all of the following:
- Distinctive: line 80
no target-project configuration unexpectedly replaced critical NLA fields.
- Distinctive: line 102
Do not claim success if any required value is unknown or if the plugin and bootstrap skill were not observed.
- Distinctive: line 116
Do not restart work on the Draft 0.4 transactional installer, managed launcher, immutable catalog, or full benchmark laboratory unless the …
- Distinctive: line 121
Use focused tests proportional to the change.
- Distinctive: line 122
Never claim completion without fresh verification evidence.
- Distinctive: line 123
Do not commit or push unless the user asks for it.
- Distinctive: line 124
Do not open a pull request to Superpowers upstream for NLA-specific changes.
- Distinctive: line 126
Do not claim support for another CLI without a complete NLA end-to-end integration test.
≈ “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
Next Level Agent Repository Instructions
This repository contains Next Level Agent for OpenCode.
Purpose of This File
These instructions are for a coding agent, including OpenCode or Codex, that is helping a user install, verify, update, or develop NLA.
Codex may assist with the installation process, but the complete NLA runtime is currently supported only in OpenCode. Do not describe this file as evidence that NLA supports Codex CLI or any other coding-agent CLI.
Before Acting
- Read README.md for product architecture and role boundaries.
- Read INSTALL.md for the supported Alpha installation procedure.
- Read docs/PROJECT_STATUS_AND_USAGE.md for current limitations, telemetry, state locations, and roadmap.
- Read docs/DRAFT_0_4_IMPLEMENTATION_STATUS.md before proposing work from the historical Draft 0.4.
- Inspect the current Git status and preserve user changes.
Installation Objective
Install NLA as a stable local clone and launch OpenCode against the user's selected project with:
OPENCODE_CONFIG=/absolute/path/to/next-level-agent/opencode.json \
opencode /absolute/path/to/projectThe supported Alpha installation does not require the unfinished transactional installer described by Draft 0.4.
Installation Rules
- Ask which project the user wants to run NLA against if it is not clear from the current workspace.
- Use a stable clone path such as
~/.local/share/nla/next-level-agent. - Do not overwrite an existing clone, global OpenCode config, project config, shell profile, provider config, or credentials.
- Before updating an existing clone, run
git status --short --branchand preserve local changes. - Do not run destructive Git commands.
- Do not store API keys in the repository or NLA state.
- Do not change model bindings without showing the proposed change to the user when it affects provider access or cost.
- Do not install packages or providers unless the user authorizes it.
- Prefer
OPENCODE_CONFIGover copying the complete NLA configuration into every project. - Treat
opencode debug configas a required non-LLM verification step. - Use a harmless read-only smoke test before a real task.
- Report exact evidence, including OpenCode version, resolved plugin path, default agent, skills path, model bindings, and smoke-test result.
Required Preflight
Run:
opencode --version
git --version
git -C /absolute/path/to/next-level-agent status --short --branchInspect:
opencode.json
config/model-pools.jsonCheck that every enabled pool has at least one model the user can access. Public defaults use models that passed the documented development smoke tests; provider availability can still change, so verify the resolved model list during installation.
Required Verification
From the target project, run:
OPENCODE_CONFIG=/absolute/path/to/next-level-agent/opencode.json \
opencode debug configVerify all of the following:
- the Next Level Agent plugin is loaded from the intended clone;
default_agentisnla;- the NLA roles are present;
- the NLA skills directory is present;
- compaction settings are active;
- no target-project configuration unexpectedly replaced critical NLA fields.
Then start a persistent OpenCode TUI for the target project and run a harmless read-only prompt. Do not use one-shot opencode run to validate controlled compaction.
Expected Installation Report
Report:
NLA clone:
Target project:
OpenCode version:
Resolved plugin:
Default agent:
Skills path:
Primary model:
Enabled role pools:
Config conflicts:
Smoke test:
Known warnings:Do not claim success if any required value is unknown or if the plugin and bootstrap skill were not observed.
Development Boundaries
NLA Core is the current priority. Focus on:
- hard role permissions;
- Task Context Packet validation;
- workflow transition enforcement;
- Supervisor behavior tests;
- practical Tier 2 and Tier 3 end-to-end tests;
- stable model pools and deterministic failure fixtures;
- memory, compaction, restoration, and telemetry reliability.
Do not restart work on the Draft 0.4 transactional installer, managed launcher, immutable catalog, or full benchmark laboratory unless the user explicitly places NLA Managed Profile back in scope.
Repository Safety
- Preserve unrelated and pre-existing changes.
- Use focused tests proportional to the change.
- Never claim completion without fresh verification evidence.
- Do not commit or push unless the user asks for it.
- Do not open a pull request to Superpowers upstream for NLA-specific changes.
- Preserve Superpowers attribution and license boundaries.
- Do not claim support for another CLI without a complete NLA end-to-end integration test.
Does 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.

