Agent kernel

Skill imMamdouhaboammar/agent-kernel/.agents/skills/agent-kernel

Use Agent Kernel when an AGENTS.md-compatible coding agent needs shared durable rules, reviewed memory proposals, Failure Lessons, bounded context, local sessions, safe project connection, provider isolation, Architecture Guardian, hooks, MCP, or release verification. Trigger on: remember this rule, capture this failure, search past debugging evidence, connect this repo safely, prevent architecture drift, configure local MCP, secure the daemon, or make multiple coding agents follow the same reviewed workflow.From its SKILL.md

Install

npx -y skills add imMamdouhaboammar/agent-kernel --skill agent-kernel

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.9 KB, as published. Nobody here has run it

Agent Kernel for AGENTS-compatible agents

Use the canonical skill at SKILL.md for the complete operating model. This adapter gives Codex, Antigravity, and other AGENTS-compatible agents the minimum non-negotiable contract.

Start

agent-kernel --version
agent-kernel doctor
agent-kernel status

Before modifying an existing project:

agent-kernel compile
agent-kernel-safe-link . --dry-run
agent-kernel-safe-link .

Read AGENTS.md, .codex/AGENTS.md, or .agents/agents.md as applicable. Do not replace user-owned instructions outside Agent Kernel managed blocks.

Core contract

Approved source memory is canonical. Generated guidance is disposable.
Agents may capture evidence and propose memory. Users approve and publish.
Failure Lessons are evidence first; promotion creates a pending proposal.
File-backed IDs are identifiers, not paths.
MCP defaults to the bounded core tool surface. Approval is disabled by default.
The daemon is local-only by default. Remote mode requires AGENT_KERNEL_DAEMON_ALLOW_REMOTE=1 and a strong AGENT_KERNEL_DAEMON_TOKEN.
Provider targets come from reviewed project manifests, not caller overrides.
Use Architecture Guardian before non-trivial code changes.

Normal work loop

agent-kernel search "<task or error>" --explain
agent-kernel context "<task>" --files src/example.mjs --budget 1200
agent-kernel failure search "<error signature>"

Create a pending durable-memory proposal:

agent-kernel propose \
  --from codex \
  --text "<durable rule>" \
  --reason "<why it should persist>"

Capture a verified failure lesson:

agent-kernel failure capture \
  --from codex \
  --type test-failure \
  --command "npm test" \
  --exit-code 1 \
  --text "<redacted error>" \
  --root-cause "<supported cause>" \
  --fix "<verified fix>"

Architecture workflow

Load architecture-guardian before a feature, refactor, new dependency, schema change, public API change, or cross-module fix.

agent-kernel architecture doctor .
agent-kernel architecture discover . --json
agent-kernel architecture reuse "<capability>" . --json
agent-kernel architecture check . --json

Do not silently broaden policy, baseline, contract, or exception scope.

Project and runtime boundaries

agent-kernel project status --json
agent-kernel context current --json
agent-kernel session start --agent codex --project . --json
agent-kernel mcp test
agent-kernel daemon status --json

Do not expose the daemon publicly. Do not enable MCP approval, bypass mode, strict architecture hooks, replacement imports, or production provider approvals without explicit user intent.

Validation

For repository changes, run the targeted test first, then the required full gate. For Agent Kernel itself:

npm run docs:check
npm run lint
npm run typecheck
npm test

Use docs/COMMAND_REFERENCE.md, docs/AGENT_RUNBOOK.md, and docs/SECURE_RUNTIME_AND_RELEASES.md.