Dev & Engineering sdlc-automationprd-to-codeloki-clicode-reviewquality-gatesmulti-providerspec-driven-developmentbackground-sessions

Loki Mode Autonomous SDLC Skill

Hand a PRD or requirements doc to the Loki CLI and let an autonomous SDLC agent take your spec all the way to deployment with minimal human intervention.

FollowSkills review · FSRS-1.0
Use with care
53/ 100 5-point scale 2.7 / 5
This review was completed under FSRS 1.0. Dimension scores are not converted; it is queued for re-review under FSRS 2.0.
1 2 3 4 5 6
1Utility12 / 20 · 3.0/5
2Reliability10 / 20 · 2.5/5
3Safety13 / 25 · 2.6/5

The OpenClaw-platform launcher for loki-mode, using --yes to skip confirmation prompts and running in the background; it does add a cost-control budget cap (--budget), but overall continues the main skill's skip-confirmation, run-autonomously pattern — the same risk considerations apply.

4Evidence5 / 15 · 1.7/5
5Usability7 / 10 · 3.5/5
6Maintenance6 / 10 · 3.0/5
Evidence confidence:Low Reviewed Jul 17, 2026
Review evidence [1]
See the full review method →

What does this skill do, and when should you use it?

This is the OpenClaw skill wrapper for Loki Mode, located at integrations/openclaw/SKILL.md, enabling an agent to launch, monitor, and control autonomous Loki SDLC sessions from a chat channel. Loki takes a spec — a PRD, GitHub issue, OpenAPI/YAML file, or one-line brief — and runs Reason-Act-Reflect-Verify cycles until the build passes its quality gates; it does not accept 'done' on an empty diff or failing tests. The skill itself writes no code; it drives the loki CLI via the bash tool: starting background sessions, polling status every 30 seconds, reporting phase progress and cost, and summarizing commits at completion. It suits users who have (or will install) the loki CLI plus at least one AI coding-agent CLI.

The skill instructs the agent to: launch an autonomous SDLC session in the background with loki start <prd-path> --bg --yes --no-dashboard; poll loki status -- every 30 seconds to read the current phase (BOOTSTRAP through DEPLOYMENT), iteration number, task counts, and elapsed time; track budget via .loki/metrics/budget. (budget_limit/budget_used); control the session with loki pause/resume/stop/logs; report phase transitions, task progress (e.g. 12/20 complete), and estimated cost to the channel after each poll; and, when the session completes, run loki status -- and git log --oneline -20 for a final summary, including .loki/council/report.md if a council verdict exists.

  1. An individual developer wants an agent to autonomously build a feature from a PRD while only watching progress reports in the chat channel
  2. A team needs a full SDLC cycle (discovery, architecture, development, QA, deployment) run in the background with periodic phase-transition reports
  3. A user kicks off a multi-round bug-fixing campaign and wants cost capped via --budget with automatic pause when the limit is exceeded
  4. A user says 'loki mode' or provides a requirements document, expecting autonomous execution to trigger without typing CLI commands manually
  5. Running cost-sensitive long builds where every progress update must include estimated spend (e.g. $4.50 / $50.00)

What are this skill's strengths and limitations?

Pros
  • True autonomous execution: background mode lets the session outlive the tool call, with --yes skipping confirmation prompts
  • Built-in cost control: --budget sets a USD cap with auto-pause, and cost is included in every progress update
  • Structured status output: -- exposes phase, task counts, PID, elapsed time, and more for reliable polling
  • Honest completion mechanism: Loki's quality gates and Evidence Receipt refuse 'done' claims on empty diffs or failing tests
  • Multi-provider: choose between Claude, Codex, and Gemini CLI
Limitations
  • Heavy dependencies: requires the loki CLI plus at least one AI coding-agent CLI with its API key — all mandatory
  • Test quality of generated code depends on AI assertions; complex domain logic still needs human review
  • Does not actually deploy: Loki prints the deploy command but never runs a cloud CLI or git push — deployment is manual
  • Per the README, the Gemini CLI provider was deprecated in v7.5.18, so the gemini option mentioned in the skill docs may be unavailable
  • Polling-based monitoring (every 30 seconds) rather than real-time push — clunky for short tasks
  • Licensed under BUSL-1.1 (source-available), not OSI open source; commercial use needs license review

How do you install this skill?

Prerequisites: 1) Install the loki CLI on the host (npm install -g loki-mode or Homebrew: brew tap asklokesh/tap && brew install loki-mode; Bun and Docker also work); 2) Install one of Claude Code, Codex CLI, or Gemini CLI; 3) Set the corresponding API key (ANTHROPIC_API_KEY, OPENAI_API_KEY, or GOOGLE_API_KEY). The skill file lives at integrations/openclaw/SKILL.md in the repo; place it in an Agent Skills-compatible skills directory to be discovered. Run loki doctor to verify your environment. The source does not document the specific install directory for the OpenClaw platform.

How do you use this skill?

In chat, have the agent: 1) Trigger by saying 'build/implement/develop a feature from a PRD', providing a requirements document, saying 'loki mode', or asking to run a full SDLC cycle on a codebase; 2) Start with bash(command: "loki start <prd-path> --bg --yes --no-dashboard", pty: true, background: true, workdir: "<project-dir>"), optionally adding --provider <claude|codex|gemini> and --budget <amount> (USD limit, auto-pauses); 3) Monitor: the agent runs loki status -- every 30 seconds and reports phase, task counts, elapsed time, and cost; 4) Control: loki pause/resume/stop/logs; 5) Once status is stopped/completed, review the final summary, git commits, and council report.

How does this skill compare with similar options?

The README compares Loki Mode with bolt.new, Replit, and Lovable, claiming it is the only one that is fully self-hosted, source-available (BUSL-1.1), and includes automated quality verification: the competitors are not self-hosted, don't let you use your own keys, and lack the 8 quality gates and blind code review. Note this skill is only a wrapper that drives the CLI, so it is not directly comparable to those hosted app-generation platforms.

FAQ

How much does a build cost?
README estimates range from roughly 5-30 minutes for simple projects to 2+ hours for complex ones, but actual cost depends on provider pricing and project complexity. The skill supports `--budget <amount>` for a USD cap with auto-pause, and `loki plan` gives a cost estimate before execution.
Does it really run unattended and deploy the app?
It runs the SDLC loop autonomously until verified completion (backed by Evidence Receipts and quality gates), but it does not deploy: `loki deploy` only prints the deploy command, and Loki never runs a cloud CLI or git push — deployment is a human step.
What if the session crashes or hangs?
The skill explicitly requires checking `loki logs --tail 50` before restarting after a crash; status is polled via `loki status --`, whose status field covers inactive/running/paused/stopped/completed/unknown.
Do I have to use Claude? How good are other providers?
The skill supports --provider across claude/codex/gemini. However, the README shows only Claude Code is the E2E-verified Tier 1 provider; Codex and Aider are experimental (sequential, no parallel agents), and Gemini CLI was deprecated in v7.5.18.

More skills from this repository

All from asklokesh/loki-mode

Related skills