Skip to main content

Agent Intel

kindling intel keeps your coding agent in sync with how kindling works. It writes a context file into the agent's system prompt location so the agent knows about your cluster, CLI commands, dependency injection, build protocol, and secrets flow — without you having to explain it every time.

How it works​

Kindling supports four coding agents out of the box:

AgentConfig file
GitHub Copilot.github/copilot-instructions.md
Claude CodeCLAUDE.md
Cursor.cursor/rules/kindling.mdc
Windsurf.windsurfrules

When intel activates, kindling:

  1. Backs up any existing agent config file (stored alongside the original)
  2. Writes a context file that includes CLI commands, architectural principles, dependency auto-injection tables, Kaniko compatibility notes, and project-specific details (detected languages, Dockerfiles, CI status)
  3. Tracks state in .kindling/intel-state.json

When intel deactivates, the originals are restored exactly as they were.

Dashboard

You can toggle Agent Intel from the dashboard home page with one click, or press ⌘K and type "intel". See Dashboard for details.

Auto-lifecycle​

Intel is designed to be hands-free. It activates and deactivates automatically based on your kindling usage:

  • Any kindling command activates intel if it isn't already active (and hasn't been manually disabled)
  • After 1 hour of inactivity (no kindling commands), the next invocation restores originals before re-activating with a fresh context
  • kindling intel off disables auto-activation entirely — intel stays off until you explicitly run kindling intel on

This means your agent always has up-to-date context while you're actively developing, and your config files are clean when you're not.

Manual control​

Activate​

kindling intel on

Activates intel immediately. If it was previously disabled with kindling intel off, this clears the disable flag and re-enables auto-lifecycle.

Deactivate​

kindling intel off

Restores all original agent config files and sets a disable flag so auto-lifecycle won't re-activate until you explicitly run kindling intel on.

Check status​

kindling intel status

Shows whether intel is active, which agent files were written, and when the last interaction occurred.

What the context file contains​

The generated context includes:

  • Architectural principles — deploy with kindling deploy, builds use Kaniko, dependencies go in the DSE YAML, secrets go through kindling secrets set
  • Dependency auto-injection table — which env vars are injected for each dependency type (e.g. postgres → DATABASE_URL)
  • CLI reference — every command with a one-line description
  • Key files — where workflows, environment specs, and context files live
  • Secrets flow — kindling secrets set → K8s Secret → secretKeyRef
  • Build protocol — source tarball → Kaniko → localhost:5001 → deploy
  • Kaniko compatibility notes — no BuildKit ARGs, no .git directory, Poetry needs --no-root, npm needs cache redirect
  • Project-specific details — detected languages, Dockerfiles found, CI workflow status

Dashboard integration​

The web dashboard includes an Agent Intel card on the overview page. From the browser you can:

  • See whether intel is active or inactive
  • Toggle intel on/off with a single click
  • View which agent config files are managed

The dashboard also has a Generate Workflow command in the command menu (⌘K) that streams AI workflow generation output in real time.

Files created​

FilePurpose
.kindling/intel-state.jsonTracks active state, backups, written files, last interaction
.kindling/intel-disabledPresent when auto-lifecycle is disabled
*.kindling-backupBackup of the original agent config (alongside the original)
tip

Add .kindling/ to your .gitignore — it's local state that shouldn't be committed. The agent config files themselves (like .github/copilot-instructions.md) are committed, so your team benefits from the context too.

Examples​

# Activate intel for all detected agents
kindling intel on

# Check what's active
kindling intel status

# Disable and restore originals
kindling intel off

# Intel also activates automatically with any command:
kindling status    # ← intel activates in the background
kindling deploy    # ← context refreshes automatically