agentleFS
ww-w-aiww-w-ai/bkit-gemini/skills/control/SKILL.md

control

Skill66 starsChanged 7 months agoApache-2.0View on GitHub

View and change bkit automation settings. Shows current automation level, feature flags, trust score, and guardrail configuration.

  • Reads credentials
  • Deletes or force-pushes
  • Commits and pushes

8 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 0 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 7 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read .bkit/config.json for current settings

    line 90
  2. Very specific:

    Update the automationLevel field

    line 102
  3. Very specific:

    Write back to .bkit/config.json using write_file

    line 103

“Common” is measured: kept by at least 6 in 10 of the 7 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Control Skill

View and manage bkit automation levels and guardrail settings

Commands

CommandDescriptionExample
/control statusShow current automation state/control status
/control level L0Set automation to manual/control level L0
/control level L1Set automation to semi-auto/control level L1
/control level L2Set automation to auto-confirm/control level L2
/control level L3Set automation to full-auto (read-only)/control level L3
/control level L4Set automation to autonomous/control level L4
/control stopEmergency stop (immediate L0 + halt)/control stop
/control rollbackRollback last N decisions/control rollback 5

Automation Levels (v2.0.7+ L3/L4 unlocked via bkit hook automation channel)

LevelNameBehavior
L0ManualAgent asks before every action. No automatic file writes. Emergency-stop default.
L1Semi-AutoAgent executes known-safe operations. Asks for destructive actions.
L2Auto-ConfirmAgent executes all operations. Confirms only for irreversible actions. Default for new bkit projects.
L3Full-AutoRead-only tools (Read/Glob/Grep/list_directory/web_fetch) auto-allowed via bkit before-tool hook. Destructive ops always asked. Requires trustScore >= 60.
L4AutonomousAll non-destructive tools auto-allowed. PDCA phase transitions automatic (via before-phase hook). Destructive ops always hard-denied (D6, 16 patterns matrix). Requires trustScore >= 80 + no recent rejections + no active 24h cooldown.

Hard Deny (영구 차단, L4 무관)

다음 16 destructive 패턴은 어떤 level에서도 항상 차단됩니다 (cmd-parser AST 분석): rm -rf, git push --force, drop table, truncate, delete from (no WHERE), sudo, chmod 777, /etc/* write, ~/.ssh/* write, gpg --delete-key, .env write, id_rsa* / *.pem / *.p12 write/delete, npm publish --force, git config --unset credential.*, rm .git, drop database.

Auto-Downgrade (안전망)

  • 5 consecutive rejections → L4/L3 → L2 + 24h cooldown
  • Hard deny detected (destructive op) → L4 → L3 (no cooldown)
  • Emergency stop (Ctrl+C 2회 또는 /control stop) → 즉시 L0

Gemini CLI 호환성 (v2.0.7 Wave 0 RCA로 입증)

  • Non-interactive gemini -p 모드: 'allow' decision suppress 정상 작동 ✓
  • Interactive mode: CC issue #52822와 유사 가능성, Track B PR 검증 중
  • 모든 결정은 .bkit/state/audit/{date}/decisions.jsonl에 영구 기록 (90d retention)

How to Execute

Show Status

  1. Read .bkit/config.json for current settings
  2. Display:
    • Current automation level (L0-L2)
    • Active feature flags
    • Trust score (if tracked)
    • Hook configuration summary
    • Guardrail rules in effect

Change Level

  1. Read .bkit/config.json
  2. Validate the requested level (L0, L1, or L2 only)
  3. Update the automationLevel field
  4. Write back to .bkit/config.json using write_file
  5. Confirm the change and explain what it means

Configuration File

Settings are stored in .bkit/config.json:

{
  "automationLevel": "L1",
  "features": {
    "auditLog": true,
    "checkpointOnDeploy": true,
    "autoIterate": false,
    "maxIterations": 5
  },
  "guardrails": {
    "denyPatterns": [".env", "credentials", "secret"],
    "maxFileSize": 10000,
    "requireCheckpoint": true
  }
}

Feature Flags

FlagDefaultDescription
auditLogtrueEnable audit trail logging
checkpointOnDeploytrueAuto-create checkpoint before deploy
autoIteratefalseAuto-run iterate when match rate < 90%
maxIterations5Maximum PDCA iteration cycles

Guardrail Rules

Guardrails are enforced by Extension hooks (before-model, before-agent):

  1. File deny patterns: Block writes to sensitive files
  2. Max file size: Prevent accidentally large file writes
  3. Checkpoint requirement: Require checkpoint before destructive operations
  4. Review requirement: Require human review for production changes

Output Format

## bkit Control Status

| Setting | Value |
|---------|-------|
| Automation Level | L1 (Semi-Auto) |
| Audit Logging | Enabled |
| Checkpoint on Deploy | Enabled |
| Auto-Iterate | Disabled |
| Max Iterations | 5 |

### Active Guardrails
- File deny patterns: .env, credentials, secret
- Max file size: 10,000 lines
- Checkpoint required: Yes

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.