autonomous

A command for running the complete unattended DevRails workflow, from an approved product brief or project description to a sequential queue of coding tasks. DevRails is the surrounding workflow system used to organize and execute that work.

In plain words
What is it for?
Use it to start an authorized end-to-end automated development run from a product brief, PRD, change request, or clarified project state.
Why use it?
It checks that required project memory, workflow rules, input, and execution support exist before starting. It stops when required policy files or safe operating conditions are missing.

Command

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add commands/nicelight/devrails-26/autonomous
Clone the repo
git clone --depth 1 https://github.com/nicelight/DevRails-26
Per session 19 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 3,336 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00019 $0.03336
Opus 5 $0.00010 $0.01668
Sonnet 5 $0.00004 $0.00667
Haiku 4.5 $0.00002 $0.00334

Measured 2d ago against content hash 20cbe9c9392f, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

autonomous scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/_shared/references/commands/autonomous.md · 295 lines

How it starts

The opening of the file, as written. The whole thing — 295 lines — stays where its author put it; the contents beside it link to each section on GitHub.

/autonomous - End-to-end unattended run

<input_contract> Use when an explicit Product Brief, PRD/delta, or existing clarified project state authorizes unattended work. Use /autopilot instead when the reviewed product JSON queue already exists and the Foundation gate is already closed or truthfully not_required.

Preflight:

  • .memory-bank/ exists, otherwise route /mb-init;
  • authoritative PRD/brief/delta input exists;
  • .memory-bank/workflows/{autonomy-policy,tier-policy,execute-loop,mb-sync}.md exist;
  • an executor capable of isolated child contexts is available;
  • repository safety policy permits code edits but no unapproved marketplace install, deployment, production write, secret read, or out-of-repo infrastructure change.

If a required workflow is missing, return HALT_POLICY_VIOLATION in the command response only, before creating/reusing the run protocol or making any other durable write. Name every missing path in the reason and state that no run protocol was started or changed; leave any existing .protocols/AUTONOMOUS-RUN/* untouched. The repair owner is the external DevRails installer: from an available DevRails checkout run node scripts/install-framework.mjs --bootstrap --target <project-path> --yes --sync, then resume /autonomous. Do not copy missing workflow rules inline.

If isolated child contexts are unavailable, return HALT_POLICY_VIOLATION with the next feature's fresh tasking, review, and /autonomous resume route. Do not create or change the run protocol or run tasking in the parent context.

If substantial code already exists, require/update the brownfield baseline via /map-codebase before applying the PRD delta.

Create/reuse:

  • .protocols/AUTONOMOUS-RUN/plan.md;
  • .protocols/AUTONOMOUS-RUN/status.md using the durable run checkpoint contract in .memory-bank/workflows/autonomy-policy.md#durable-run-checkpoint;
  • .protocols/AUTONOMOUS-RUN/decision-log.md;
  • .tasks/TASK-AUTONOMOUS/.

Record run metadata, scheduler mode, review coverage, blocking decisions, applied authoritative policies, queue summary linked to JSON records, failure budget, and terminal state/reason. The run protocol is resumable orchestration state, not a second task registry. </input_contract>

<hard_invariants>

  • Follow tier policy #tier-obligations and #closure-authority; load #hard-write-boundary only for a task/parallel boundary and #tier-classification-and-escalation only for a tier gap. Follow autonomy policy #phase-ownership, #experimental-parallel-execution when opted in, #durable-run-checkpoint, #required-gates, #failure-budgets, #scheduler-failure-handling, #terminal-fallback, #run-state, and #terminal-states, plus execute-loop #planning-redesign-boundary, #fresh-feature-tasking-boundary, and the applicable mb-sync contract.
  • Canonical queue execution is sequential.
  • --experimental-parallel remains opt-in and uses only existing autonomy-policy isolation rules. Never infer independence from advisory touched_files.
  • Queue/task metadata comes only from indexed JSON task records. Preserve task schema, IDs, lifecycle planned|ready|in_progress|blocked|done|done_for_prod|failed, tier, waves, Foundation dependencies, and hard runtime scopes.
  • During the Foundation phase, /autonomous owns promotion, selection, and final lifecycle decisions only for indexed feature: "FT-000" records. After the Foundation gate is done, /autopilot owns those scheduler decisions only for product records. /exe owns ready -> in_progress for the concrete task selected by either scheduler. Neither phase may adopt or mutate the other phase's tasks.
  • /autonomous must not restage /autopilot's product-queue algorithm or reinterpret its task transitions. /exe, /verify, /red-verify, and /mb-sync keep the ownership defined by tier policy #closure-authority.
  • The active scheduler writes every task closure/failure/blocking decision, status, and evidence link to the authoritative .task.json before any sync boundary.
  • /mb-sync reconciles already-written state once per wave unless an explicit current-wave durable-state dependency requires an early sync; it never chooses closure or promotion.
  • Required reviews, lint, doctor, and tier-policy.md#tier-obligations gates are not bypassed.
  • Preserve failure budgets and all existing terminal states; do not add a scheduler, status, lifecycle, or assumption registry. </hard_invariants>

Read the full file on GitHub · 295 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 295 lines · 19 tokens per session scan A 20cbe9c9392f

Subscribe to this mod's changes

autonomous is a command published in the GitHub repository nicelight/DevRails-26 (22 stars, last pushed 4d ago), licensed MIT. It adds 19 tokens to every session and 3,336 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.