# ADLC: The Agentic Development Lifecycle

> A software lifecycle built around how AI models actually fail: machine-checkable gates, auditable evidence, and native support for coding agents. The canonical theory series lives at https://voodootikigod.com/series/adlc; the toolkit is open source at https://github.com/voodootikigod/adlc.

## Site

- [ADLC: The Agentic Development Lifecycle](https://www.agenticlifecycle.ai): A software lifecycle built around how AI models actually fail, with deterministic evidence and explicit human attestation.
- [The Lifecycle: Phases and Gates](https://www.agenticlifecycle.ai/lifecycle): The eight phases P0-P7, the gate that ends each one, and the three dials set per ticket.
- [Failure Modes: Why Agents Fail](https://www.agenticlifecycle.ai/failure-modes): The eight model failure modes F1-F8 and the specific gate that defends against each.
- [ADLC vs SDLC](https://www.agenticlifecycle.ai/vs-sdlc): What transfers from 60 years of SDLC practice and what has to be rebuilt for model failure modes.
- [The Toolkit](https://www.agenticlifecycle.ai/toolkit): Zero-dependency CLIs that each enforce one machine-checkable gate, grouped by lifecycle phase.
- [Integrations: Native to Your Agent](https://www.agenticlifecycle.ai/integrations): Install the ADLC natively in Claude Code, Codex, Cursor, OpenCode, Pi, or Google Gemini.
- [ADLC for Enterprise](https://www.agenticlifecycle.ai/enterprise): Rolling out agentic development with an audit trail: gate evidence, rollout phases, and compliance framing.

## Integrations

- [ADLC for Claude Code](https://www.agenticlifecycle.ai/integrations/claude-code): Phase-routing skill, slash-command gates, rails-guard and build-gate hooks, and a multi-lens prosecutor panel. Claude is the model via --prompt-only.
- [ADLC for Codex](https://www.agenticlifecycle.ai/integrations/codex): A native Codex plugin with phase routing, frozen-rail feedback, gate tools, and project review agents.
- [ADLC for Cursor](https://www.agenticlifecycle.ai/integrations/cursor): Marketplace plugin with sessionStart context, preToolUse rails, packaged prosecutor agents + one-root MCP channel live proven on Cursor 3.20.10; rebind and live multi-root ambiguity not proven, /adlc-* commands, and CI as the backstop.
- [ADLC for OpenCode](https://www.agenticlifecycle.ai/integrations/opencode): Enforce-by-default rails-guard, full /adlc-* lifecycle commands, native adlc_gate/adlc_prosecute tools, and a multi-lens prosecution loop with a verifier.
- [ADLC for Pi](https://www.agenticlifecycle.ai/integrations/pi): Proactive tool_call blocking and reactive snapshot revert, five phase skills, native adlc_gate/adlc_prosecute tools, and the strongest team-install story among the harnesses.
- [ADLC for Google Gemini (Antigravity & JetSki)](https://www.agenticlifecycle.ai/integrations/gemini): Advisory PreToolUse rails-guard for agy, plus skills and a prosecutor agent. The unbypassable CI rail-freeze gate is the real control.
- [ADLC for GitHub Copilot](https://www.agenticlifecycle.ai/integrations/copilot): A native Copilot CLI plugin: seven lifecycle hooks, six phase skills, allowlisted MCP gate tools, and six read-only prosecution agents.

## For agents

- [Agent onboarding guide](https://www.agenticlifecycle.ai/agent-guide.md): step-by-step install and setup instructions written for a coding agent acting on a human's behalf — harness detection, install paths, first-run walkthrough, and the constraints an agent must respect.

## Documentation
# Docs

- [ADLC Documentation](/docs): The Agentic Development Lifecycle: a toolkit and harness for shipping reliable software with AI models.
- [Getting started](https://www.agenticlifecycle.ai/docs/getting-started): Install the ADLC toolkit, bootstrap a repo, and run your first gate in under five minutes.
- Theory & Introduction
  - [Theory & Introduction](https://www.agenticlifecycle.ai/docs/theory): How the ADLC phases and model failure modes map to the toolkit, with deep links to the original series.
- [The Lifecycle](https://www.agenticlifecycle.ai/docs/lifecycle): The ADLC phase and gate map, and which toolkit CLI enforces each gate.
- Toolkit
  - [Toolkit](https://www.agenticlifecycle.ai/docs/toolkit): The ADLC toolkit: one deterministic gate per lifecycle phase, installed once through @adlc/cli.

  - **Spec & ticket shaping**
  - [tickets](https://www.agenticlifecycle.ai/docs/toolkit/tickets): Canonical immutable ticket snapshots, sharded storage, migration, archive, transactions, and provenance.
  - [parallax](https://www.agenticlifecycle.ai/docs/toolkit/parallax): Fans N independent readers over a request and surfaces only the divergences as questions humans must answer.
  - [spec-lint](https://www.agenticlifecycle.ai/docs/toolkit/spec-lint): Audits a spec for acceptance criteria that lack a concrete verification method.
  - [premortem](https://www.agenticlifecycle.ai/docs/toolkit/premortem): Tells a frontier model the project already failed and asks for the postmortem, which surfaces concrete risks anchored to the spec.
  - [coldstart](https://www.agenticlifecycle.ai/docs/toolkit/coldstart): A cheap-tier LLM plays a fresh agent and lists every human-only question a ticket leaves unanswered before the build queue.
  - [ticket-sync](https://www.agenticlifecycle.ai/docs/toolkit/ticket-sync): Two-way sync between ADLC tickets and an external tracker (GitHub Issues), dry-run by default and failing closed on conflict.
  - [ticket-prune](https://www.agenticlifecycle.ai/docs/toolkit/ticket-prune): Reports and archives stale, already-shipped tickets out of .adlc/tickets.json so live work is not buried under leftovers.

  - **Execution supervision & rails**
  - [preflight](https://www.agenticlifecycle.ai/docs/toolkit/preflight): Batches every environment and permission check the agent fleet needs into one pass before fan-out, so failures front-load instead of stalling mid-flight.
  - [model-router](https://www.agenticlifecycle.ai/docs/toolkit/model-router): Deterministically assigns every ticket a model tier and ladder mode from the ticket DAG and ledger history, with no LLM calls.
  - [quartermaster](https://www.agenticlifecycle.ai/docs/toolkit/quartermaster): The operator-local supply layer. Decides which model runs which lifecycle job from a registry outside the repo, and routes build work by derived class rather than by the caller's label.
  - [merge-forecast](https://www.agenticlifecycle.ai/docs/toolkit/merge-forecast): Predicts merge conflicts before any agent runs, certifies the safe fan-out width, and emits a foundation-first merge schedule.
  - [build-gate](https://www.agenticlifecycle.ai/docs/toolkit/build-gate): Denies starting a high-risk ticket's build in a degraded (context-rot) session unless an audited override is durably recorded.
  - [context-handoff](https://www.agenticlifecycle.ai/docs/toolkit/context-handoff): Absolute context-band signals, session-terminal mutation deny (D1–D3), and the operator CLI for write/resume/bypass/repair/unlock — F3 continuity handoff.
  - [rails-guard](https://www.agenticlifecycle.ai/docs/toolkit/rails-guard): Blocks builder edits to frozen rail paths and greps added diff lines for undeclared suppression markers.
  - [flail-detector](https://www.agenticlifecycle.ai/docs/toolkit/flail-detector): Analyzes a build-session log for flail signatures and fails the gate when an agent is spinning.
  - [consensus-fix](https://www.agenticlifecycle.ai/docs/toolkit/consensus-fix): Fans N independent LLM fixes at a failing test, gates each against the rails, and picks the smallest consensus winner.
  - [runner](https://www.agenticlifecycle.ai/docs/toolkit/runner): Asserts that each ADLC phase's required manifest evidence is present, and fails closed when reviewed content changes.
  - [fleet](https://www.agenticlifecycle.ai/docs/toolkit/fleet): Parallel ticket orchestration. Dispatches ready tickets to sandboxed Claude Code workers in isolated worktrees, gates and prosecutes each, and merges to a per-run integration branch with deterministic control flow.
  - [autopilot](https://www.agenticlifecycle.ai/docs/toolkit/autopilot): Quota-gated local issue-to-PR loop. Picks the next GitHub issue, shapes an ADLC ticket, dispatches one sandboxed fleet run per issue, runs the outer gates, attests with a cross-model review and opens the PR. P6 (merge) stays human.

  - **Review evidence & calibration**
  - [behavior-diff](https://www.agenticlifecycle.ai/docs/toolkit/behavior-diff): Captures each route's observable HTTP behavior before and after a change, then diffs in behavior-space so the human reviews six items, not a 5,000-line diff.
  - [gate-manifest](https://www.agenticlifecycle.ai/docs/toolkit/gate-manifest): A hash-chained evidence ledger recording what each ADLC gate verified, with optional HMAC signing so the chain attests authorship, not just consistency.
  - [hollow-test](https://www.agenticlifecycle.ai/docs/toolkit/hollow-test): Diff-scoped mutation gate. Mutates only the lines your change touched and fails if any mutant survives your test suite, which proves the coverage is load-bearing.
  - [prosecute](https://www.agenticlifecycle.ai/docs/toolkit/prosecute): A P5 review-evidence recorder. It validates, hashes, and records ticket- and revision-bound reviewer findings, and passes only after two dry passes across three lenses.
  - [review-calibration](https://www.agenticlifecycle.ai/docs/toolkit/review-calibration): Measures reviewer recall by planting bugs and scoring whether the review command catches them.
  - [model-ratchet](https://www.agenticlifecycle.ai/docs/toolkit/model-ratchet): Ranks hot files (churn × import criticality) for re-prosecution after a model or repo drift.
  - [gate-fuzzing](https://www.agenticlifecycle.ai/docs/toolkit/gate-fuzzing): Fans model instances to generate wrong-but-passing diffs and reports any gate they defeat.

  - **Compounding defenses**
  - [lesson-foundry](https://www.agenticlifecycle.ai/docs/toolkit/lesson-foundry): Clusters recurring prosecution findings and routes each to its cheapest permanent defense (lint, skill, or spec gap).
  - [rejection-mining](https://www.agenticlifecycle.ai/docs/toolkit/rejection-mining): Mines human PR review objections into reusable prosecution lenses so recurring "no"s are caught pre-submit.
  - [skill-rot](https://www.agenticlifecycle.ai/docs/toolkit/skill-rot): Verifies the checkable claims inside SKILL.md files still hold against the current repo and stamps fresh ones.
  - [backlog-groom](https://www.agenticlifecycle.ai/docs/toolkit/backlog-groom): Verifies each open issue's premise against the code at HEAD, clusters and ranks the backlog, and emits a versioned groomed set. Read-only.

  - **Shared foundation**
  - [cli](https://www.agenticlifecycle.ai/docs/toolkit/cli): The umbrella dispatcher: one stable command surface that routes every adlc verb to the right gate in the suite.
  - [core](https://www.agenticlifecycle.ai/docs/toolkit/core): The frozen shared foundation every ADLC tool imports: exit-code helpers, provider selection, and the ledger, ticket, rail, and mutation primitives.
- Integrations
  - [Integrations](https://www.agenticlifecycle.ai/docs/integrations): Native ADLC integrations for Google Gemini (Antigravity/JetSki), Claude Code, Codex, Cursor, opencode, and pi.
  - [Claude Code](https://www.agenticlifecycle.ai/docs/integrations/claude-code): Adopt the ADLC inside Claude Code with a phase-routing skill, gate commands, a prosecutor subagent, and hooks that fire the gates automatically.
  - [Codex](https://www.agenticlifecycle.ai/docs/integrations/codex): Install ADLC as a native Codex plugin with skills, lifecycle hooks, MCP gate tools, and project-scoped agents.
  - [Cursor](https://www.agenticlifecycle.ai/docs/integrations/cursor): Native ADLC integration for Cursor: marketplace plugin with sessionStart context, preToolUse rails, packaged prosecutor agents + MCP one-root channel (rebind unverified), /adlc-* commands, and CI as the backstop.
  - [opencode](https://www.agenticlifecycle.ai/docs/integrations/opencode): Native ADLC integration for the opencode terminal agent: an enforce-by-default rails-guard hook, nine lifecycle commands, native adlc_gate/adlc_prosecute tools, and a deterministic P5 prosecution loop (5 lenses + verifier) plus a prosecutor meta-agent.
  - [pi](https://www.agenticlifecycle.ai/docs/integrations/pi): Native ADLC extension for the pi coding agent: in-session rail, scope, and suppression enforcement with git-diff revert, plus five phase skills.
  - [Google Gemini (Antigravity & JetSki)](https://www.agenticlifecycle.ai/docs/integrations/gemini): Native ADLC integration for Google Gemini (Antigravity/JetSki): an in-session rails-guard hook plus the unbypassable CI diff gate that is the real enforcement layer.
  - [GitHub Copilot](https://www.agenticlifecycle.ai/docs/integrations/copilot): Native ADLC integration for the GitHub Copilot CLI: seven lifecycle hooks, six phase skills, allowlisted MCP gate tools, and six read-only prosecution agents, backed by the unbypassable CI rail gate.
- Reference
  - [Reference](https://www.agenticlifecycle.ai/docs/reference): The contracts every ADLC tool honors: conventions, exit codes, the .adlc/ runtime, and the architecture decision records.
  - [Tool Conventions](https://www.agenticlifecycle.ai/docs/reference/conventions): The contract every packages/ tool follows so ~20 independently built CLIs feel like one product.
  - [Exit Codes](https://www.agenticlifecycle.ai/docs/reference/exit-codes): The canonical 0/1/2 exit-code convention that makes ADLC gates deterministic in CI.
  - [The .adlc/ Runtime](https://www.agenticlifecycle.ai/docs/reference/adlc-runtime): The on-disk state ADLC gates read and write: tickets, the active-ticket pointer, config, ledgers, evidence, and the environment variables that steer them.
  - [Recommended Models by Phase](https://www.agenticlifecycle.ai/docs/reference/models-by-phase): A dated binding from the router's abstract tiers to concrete provider models, with verification guidance for changing catalogs.
  - [Architecture Decision Records](https://www.agenticlifecycle.ai/docs/reference/adrs): The accepted ADRs behind ADLC's editor integrations and adversarial-review gates.