aide-agents-md
The AIDE pattern for session-scoped, workspace-scoped, and user-scoped agent instructions — the AGENTS.md (and equivalent) file convention.
Skill by AnonymousNomad
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/anonymousnomad-aide-agents-md
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
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: agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/aide-agents-md
What's inside
- Skill name
aide-agents-md- Description
- The AIDE pattern for session-scoped, workspace-scoped, and user-scoped agent instructions — the AGENTS.md (and equivalent) file convention. Use when implementing the AIDE analog of Claude Code's CLAUDE.md, Cursor's Rules, Aider's conventions file, or Cline's Memory Bank. Use when the user asks "make AIDE always do X" or "this project uses tabs, never spaces" or "when I say commit I mean signed". Use when reviewing the prompt composition for the chat path.
- License
- Apache-2.0
- Instructions
- 219 lines in SKILL.md · view on GitHub
Agent Skills spec issues
- files: referenced but missing: scripts/aide-bundle.cjs
Files (1)
- SKILL.md
SKILL.md (first lines)
# AGENTS.md — Persistent Agent Instructions, Scoped Three Ways
Born 2026-08-31 from the wiring audit. AIDE has `harness/credo.md` (the global credo) and `aide-advanced-orchestration` Pattern 5 (AGENTS.md export) but the **load order, scoping rules, and edit UX** are not specified or wired. Every rival (Claude Code, Cursor, Cline, Copilot, Aider) has this. AIDE's version is "the credo is global, period" — too coarse. This skill IS the wire-in.
## The AIDE AGENTS.md contract (3 scopes, 1 precedence)
```
Priority (highest to lowest):
1. SESSION scope: .aide/agents/session.md (ephemeral, current session only)
2. WORKSPACE scope: <workspace>/AGENTS.md (committed, project-wide)
3. USER scope: ~/.aide/agents/user.md (cross-workspace, operator's preferences)
4. GLOBAL scope: harness/credo.md (the oath, non-overridable)
```
**Why 4 levels, not 3:** the global credo is **non-overridable** by the other three. It's the oath. The other three are preferences layered on top.
**The "non-overridable" rule:** the credo defines the oaths (protect user, evidence over confidence, etc.). Workspace AGENTS.md can ADD constraints ("always run tests before commit") but cannot REMOVE an oath. A diff that conflicts with the credo is rejected with `OATH_CONFLICT`.
## Files to touch (when wiring)
| File | Change |
|---|---|
| `node/src/services/agent-instructions.mjs` | NEW: `loadInstructions({workspace, userHome, sessionId})` returns the merged, deduplicated, oath-checked instruction set. |
| `node/src/services/scaffold.mjs` | EXTEND: `composeScaffold` calls `loadInstructions` and prepends the merged set to the system prompt. |
| `node/src/routes/agent.ts` (or new `node/src/routes/instructions.ts`) | ADD: `GET /api/instructions` (show merged), `PUT /api/instructions/scope/:scope` (edit one scope), `GET /api/instructions/scopes` (list all 4). |
| `common/contracts/agent.ts` or new `common/contracts/instructions.ts` | ADD: the 4 zod schemas. |
| `node/src/server.ts` | Wire the routes. |
| `tests/arch/agents-md.test.ts` | NEW: 5 tests (scope precedence, oath conflict, dedup, session expiry, edit UX). |
| `scripts/aide-bundle.cjs` | (optional) `bundle instructions` CLI. |
| `browser/src/...` | (optional) "Edit workspace instructions" UI in the settings panel. |
## The contract (zod-strict)
```ts
// in common/contracts/instructions.ts
export const InstructionScope = z.enum(['session', 'workspace', 'user', 'global']);
export const InstructionEntry = z.object({
scope: InstructionScope,
path: z.string(), // file path for verification
body: z.string(),
lines: z.number().int().gte(0),
modified_at: z.number().int(),
oath_check: z.enum(['pass', 'fail', 'unknown']).default('unknown'),
oath_conflict: z.string().nullable().optional()
}).strict();
## The load + merge algorithm
```js
// node/src/services/agent-instructions.mjs
import { promises as fs } from 'node:fs';
import path from 'node:path';
import os from 'node:os';
const OATHS = [
'protect the user and their data',
'evidence over confidence',
'finish the procedure',
'preserve the workspace',
'carry knowledge forward'
];
const SCOPE_PATHS = {
global: 'harness/credo.md',
workspace: 'AGENTS.md',
user: path.join(os.homedir(), '.aide', 'agents', 'user.md'),
session: path.join('.aide', 'agents', 'session.md')
};
export async function loadInstructions({ workspace, sessionId }) {
const entries = [];
const conflicts = [];
for (const [scope, relPath] of Object.entries(SCOPE_PATHS)) {
const absPath = scope === 'user' || scope === 'global' ? relPath : path.join(workspace, relPath);
let body;
try { body = await fs.readFile(absPath, 'utf8'); }
catch (e) {
if (e.code === 'ENOENT') continue;
throw e;
}
const oathCheck = checkOaths(scope, body);
const entry = {
scope, path: absPath, body,
lines: body.split('\n').length,
modified_at: (await fs.stat(absPath)).mtimeMs,
oath_check: oathCheck.status,
oath_conflict: oathCheck.conflict
};
entries.push(entry);
if (oathCheck.status === 'fail') {
conflicts.push({ scope, path: absPath, snippet: oathCheck.snippet, conflicting_oath: oathCheck.oath });
}
}
const merged = entries
.sort((a, b) => orderByScope(a.scope) - orderByScope(b.scope))
.map(e => '# [' + e.scope + ']\n' + e.body)
.join('\n\n');
return { merged, entries, conflicts };
}
function orderByScope(scope) {
return { global: 0, user: 1, workspace: 2, session: 3 }[scope];
}
function checkOaths(scope, body) {
const violations = [
## The scaffold integration
```js
// in harness/scaffold.mjs:
import { loadInstructions } from '../node/src/services/agent-instructions.mjs';
export async function composeScaffoldWithInstructions({ workspace, effectiveContextTokens, taskFamily, sessionId }) {
const baseScaffold = composeScaffold({ effectiveContextTokens, taskFamily });
const instructions = await loadInstructions({ workspace, sessionId });
const header = '[AIDE instructions | ' + instructions.entries.length + ' scope(s) loaded]';From AnonymousNomad/covert-coder, Apache-2.0.
More from AnonymousNomad/covert-coder
- 0.624score
aide-architect-editor-pattern
AnonymousNomad
The two-call decomposition for any agentic surface that needs higher quality from a smaller model — Architect reasons + plans (no file changes, no tools)…
SkillBFreshness: B★ 53Claude Code +9 - 0.624score
aide-backend-autoselect
AnonymousNomad
Backend auto-detection and selection SOP for llama-server engines in AIDE — probes candidate binaries via --list-devices, caches capabilities, selects…
SkillBFreshness: B★ 53Claude Code +9 - 0.624score
aide-advanced-orchestration
AnonymousNomad
Advanced orchestration patterns for AIDE — multi-lens review (one model, three contexts), clarifying question engine (structured constraint-gathering before…
SkillBFreshness: B★ 53Claude Code +9 - 0.624score
aide-agent-harness-convergence
AnonymousNomad
Converge AIDE's three parallel agent runtimes (legacy daemon model-manager chat + harness/orchestrator.mjs, arch-ts model-router/chat.ts + routes/agent.ts…
SkillBFreshness: B★ 53Claude Code +9 - 0.624score
aide-android-build
AnonymousNomad
Android build integration for AIDE — Gradle wrapper detection, APK build/debug/install via task service, emulator control, logcat streaming.
SkillBFreshness: B★ 53Claude Code +9 - 0.621score
adapter-serve-routing
AnonymousNomad
Define and verify deterministic format-conditional adapter routing at inference.
SkillBFreshness: B★ 53Claude Code +9 - 0.620score
aide-browser-ide
AnonymousNomad
Architectural options for running AIDE Sovereign Workbench in the browser so visitors can use it without downloading.
SkillBFreshness: B★ 53Claude Code +9 - 0.620score
aid-dap-real-debugpy-fixture
AnonymousNomad
Wire in a REAL debugpy fixture for the AIDE DAP test batteries.
SkillBFreshness: B★ 53Claude Code +9 - 0.620score
aide-architect-editor-implementation
AnonymousNomad
The implementation plan for the architect→editor two-call decomposition in AIDE's agent loop.
SkillBFreshness: B★ 53Claude Code +9 - 0.620score
aide-background-tasks
AnonymousNomad
The AIDE pattern for long-running, durable background tasks — a JSONL outbox on disk that survives daemon restarts, a drain worker that replays idempotently…
SkillBFreshness: B★ 53Claude Code +9 - 0.620score
aide-ci-diagnostics
AnonymousNomad
Diagnose GitHub Actions failures on AnonymousNomad/aide-sovereign-workbench WITHOUT authentication — no-auth evidence channels (::error annotations, job…
SkillBFreshness: B★ 53Claude Code +9 - 0.617score
PC management
AnonymousNomad
Owner's manual for THIS machine (ASUS GL703GM — i7-8750H / GTX 1060 / 16GB RAM / 120Hz).
SkillBFreshness: B★ 53Claude Code +9
Signals
- Stars
- 53
- Forks
- 15
- Open issues
- 2
- Last commit (this folder)
- Aug 31, 2026 (1 month ago)
- Last release
- Aug 10, 2026 (2 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- Apache-2.0
- Repository
- github.com/AnonymousNomad/covert-coder
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from AnonymousNomad
- 0.624score
aide-backend-autoselect
AnonymousNomad
Backend auto-detection and selection SOP for llama-server engines in AIDE — probes candidate binaries via --list-devices, caches capabilities, selects…
SkillBFreshness: B★ 53Claude Code +9 - 0.624score
aide-architect-editor-pattern
AnonymousNomad
The two-call decomposition for any agentic surface that needs higher quality from a smaller model — Architect reasons + plans (no file changes, no tools)…
SkillBFreshness: B★ 53Claude Code +9 - 0.624score
aide-advanced-orchestration
AnonymousNomad
Advanced orchestration patterns for AIDE — multi-lens review (one model, three contexts), clarifying question engine (structured constraint-gathering before…
SkillBFreshness: B★ 53Claude Code +9 - 0.624score
aide-agent-harness-convergence
AnonymousNomad
Converge AIDE's three parallel agent runtimes (legacy daemon model-manager chat + harness/orchestrator.mjs, arch-ts model-router/chat.ts + routes/agent.ts…
SkillBFreshness: B★ 53Claude Code +9 - 0.624score
aide-android-build
AnonymousNomad
Android build integration for AIDE — Gradle wrapper detection, APK build/debug/install via task service, emulator control, logcat streaming.
SkillBFreshness: B★ 53Claude Code +9 - 0.621score
adapter-serve-routing
AnonymousNomad
Define and verify deterministic format-conditional adapter routing at inference.
SkillBFreshness: B★ 53Claude Code +9
Cite this page
aide-agents-md skill. Agent Mart. https://agentmart.dev/skills/anonymousnomad-aide-agents-md (last updated 2026-10-24)