lateralus
AGENTS.md3 starsChanged 3 months agoMITView on GitHub
Single source of truth for all AI agents working in this repo. CLAUDE.md and GEMINI.md both point here. Lateralus is a lateral-thinking skill for AI coding agents that breaks debugging tunnel vision. It surfaces the user's end goal and solution horizon (long-term, MVP, POC, workaround), then generates goal-appropriate alternatives in two tiers: grounded-but-unlikely and wild/speculative reframes. Ships as a Copilot repo skill and direct install via install.sh / install.ps1. Do NOT edit .github/skills/ directly — those are CI-synced mirrors. .github/workflows/sync-skill.yml…
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 32,671 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 repository3 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- lateralus-brainstormSkill
- lateralusSkill
The file
We split this AGENTS.md into the rules it gives an agent
7 rules. About 0 are common, 2 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
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
- could be automatic: a setting could enforce it
- Very specific: line 71
Do NOT edit .github/skills/ directly — those are CI-synced mirrors.
- Distinctive: line 84
After merging a skill change, wait for the workflow before declaring the release complete.
- Distinctive: line 101
Don't merge them. Different audiences, different formats.
- Very specific: line 142
If settings.json can't be parsed, wiring is skipped and the file is never modified
- Very specific: line 152
Keep skill descriptions keyword-rich — the description field is how agents discover when to load the skill.
- Very specific: line 153
Keep SKILL.md under 500 lines.
- Very specific: line 155
After merging a skill edit, confirm CI synced the .github/skills/ mirror before marking done.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 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
lateralus — project context
Single source of truth for all AI agents working in this repo.
CLAUDE.mdandGEMINI.mdboth point here.
Project overview
Lateralus is a lateral-thinking skill for AI coding agents that breaks debugging tunnel vision. It surfaces the user's end goal and solution horizon (long-term, MVP, POC, workaround), then generates goal-appropriate alternatives in two tiers: grounded-but-unlikely and wild/speculative reframes.
Ships as a Copilot repo skill and direct install via install.sh / install.ps1.
What lives where
lateralus/
├── README.md # Front door (product pitch + install one-liners)
├── INSTALL.md # Full install matrix
├── CONTRIBUTING.md # Dev guide and file ownership
├── AGENTS.md # This file — maintainer context for all agents
├── CLAUDE.md / GEMINI.md # Thin stubs pointing here
│
├── install.sh # macOS / Linux / WSL installer
├── install.ps1 # Windows PowerShell installer
│
├── skills/ # ALL skills — single source of truth
│ ├── lateralus/
│ │ ├── SKILL.md # Full debugging skill
│ │ └── README.md
│ └── lateralus-brainstorm/
│ └── SKILL.md # Full brainstorming / plan-mode skill
│
├── agents/ # Subagents — single source of truth
│ ├── lateralus-ideator-ground.md # Generates Tier 1 grounded hypotheses
│ ├── lateralus-ideator-balanced.md # Generates middle-ground hypotheses
│ ├── lateralus-ideator-wild.md # Generates Tier 3 wild reframes
│ └── lateralus-workaround.md # Makeshift bypasses with debt log
│
├── hooks/ # PostToolUse hook implementations
│ ├── lateralus-hook.py # Python hook (macOS / Linux / WSL)
│ └── lateralus-hook.ps1 # PowerShell hook (Windows, requires pwsh 6+)
│
├── .github/skills/ # Copilot-discoverable mirrors (CI-synced)
│ ├── lateralus/SKILL.md
│ └── lateralus-brainstorm/SKILL.md
│
└── .github/workflows/
└── sync-skill.yml # Syncs skills/ → .github/skills/ on push to mainFile ownership — edit only these
| I want to change... | Edit this file |
| Core skill behavior (tiers, horizon routing, rules) | skills/lateralus/SKILL.md |
| Brainstorming / plan-mode skill | skills/lateralus-brainstorm/SKILL.md |
| Scout (interview + audit) | Built into skills/lateralus/SKILL.md Step 0 |
| Tier 1 grounded hypotheses | agents/lateralus-ideator-ground.md |
| Middle-ground hypotheses | agents/lateralus-ideator-balanced.md |
| Tier 3 wild reframes | agents/lateralus-ideator-wild.md |
| Makeshift bypass rules | agents/lateralus-workaround.md |
| Hook behavior | hooks/lateralus-hook.py, hooks/lateralus-hook.ps1 |
| Install steps | install.sh, install.ps1, INSTALL.md |
| Agent context (this file) | AGENTS.md — then CLAUDE.md/GEMINI.md update automatically |
Do NOT edit .github/skills/ directly — those are CI-synced mirrors.
CI sync workflow
.github/workflows/sync-skill.yml triggers on push to main when any skills/**/SKILL.md changes.
What it does:
- Copies
skills/lateralus/SKILL.md→.github/skills/lateralus/SKILL.md - Copies
skills/lateralus-brainstorm/SKILL.md→.github/skills/lateralus-brainstorm/SKILL.md - Commits with
[skip ci]to avoid loops.
After merging a skill change, wait for the workflow before declaring the release complete.
Skill system
Three skills ship from this repo:
| Skill | File | Purpose |
lateralus | skills/lateralus/SKILL.md | Full debugging skill — goal context + two-tier ideation |
lateralus-brainstorm | skills/lateralus-brainstorm/SKILL.md | Brainstorming / plan mode — Grounded, Balanced, Wild |
Each skill has:
- A
SKILL.md(LLM-facing prompt body — what the agent loads) - A
README.mdalongside for humans browsing GitHub
Don't merge them. Different audiences, different formats.
Agent system
Four subagents covering the full lateralus workflow:
| Stage | Agent | Model |
| 1. Ground | lateralus-ideator-ground | sonnet |
| 2. Balanced | lateralus-ideator-balanced | sonnet |
| 3. Wild | lateralus-ideator-wild | sonnet |
| 4. Bypass | lateralus-workaround | sonnet |
Interrogation (interview + codebase audit) is built into the skill itself as Step 0.
Horizon routing (after Step 0 context block is complete) — user picks from the menu:
- Just unblock / demo → workaround directly
- Long-term / MVP → user picks ground + wild (or agent runs both sequentially)
- POC / test → ideator-ground only
- Time-pressured → ideator-balanced only
Install system
Two plain shell scripts. Both install a PostToolUse hook and wire it into ~/.claude/settings.json.
install.sh and install.ps1:
- Copy all three skills to
~/.claude/skills/ - Copy all four agent files to
~/.claude/agents/ - Copy the hook to
~/.claude/hooks/ - Wire the hook into
~/.claude/settings.json(idempotent; backs up.bakbefore writing) - Respect
CLAUDE_CONFIG_DIRenv var - Work from curl-pipe or local clone
- Safe to re-run
Hook notes:
install.shrequires Python 3 for JSONC-tolerant settings.json mergeinstall.ps1requirespwsh(PowerShell 6+) for hook wiring; skips gracefully on PS 5.1- If
settings.jsoncan't be parsed, wiring is skipped and the file is never modified - A
.bakcopy is written alongsidesettings.jsonbefore any write
Key rules for agents working here
- Edit
skills/<name>/SKILL.mdfor behavior changes. Never edit the.github/skills/mirrors. - Edit
agents/<name>.mdfor subagent behavior. All agents are single-source at the repo root. - Edit
AGENTS.mdfor maintainer context.CLAUDE.mdandGEMINI.mdare stubs — don't edit them. - Keep skill descriptions keyword-rich — the
descriptionfield is how agents discover when to load the skill. - Keep SKILL.md under 500 lines. Use reference files if it grows.
install.shandinstall.ps1are the single source for install logic. Don't add OS-specific logic to one without the other.- After merging a skill edit, confirm CI synced the
.github/skills/mirror before marking done. - README is user-facing. Keep install one-liners accurate. If an install path changes, update README, INSTALL.md, and both install scripts.
@./skills/lateralus/SKILL.md @./skills/lateralus-brainstorm/SKILL.md @./agents/lateralus-ideator-ground.md @./agents/lateralus-ideator-balanced.md @./agents/lateralus-ideator-wild.md @./agents/lateralus-workaround.md
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.

