AgentFlow 1.0 · Command-line manual

AGENTFLOW-SDLC(1)

Install, inspect, update, and validate a reviewable delivery system around an existing repository.

NAME

agentflow-sdlc — process guardrails for AI-assisted delivery

The CLI installs and maintains AgentFlow policy, templates, hooks, validators, evidence contracts, adapters, and project-local configuration. It operates on an existing project; it does not generate an application or replace its stack.

SYNOPSIS

node bin/cli.mjs <command> --target <project> [options]

Run from the v1.0.0 source checkout. Quote paths containing spaces.

AGENTFLOW 1.0 COMMANDS

initInstall framework files and seed project-owned configuration after approval.
syncUpdate managed files while preserving project-owned divergence and ownership classifications.
doctor / doctor-envInspect installation drift, framework health, and environment readiness.
onboarding-prompt / update-promptPrint read-only-first assisted adoption or update handoffs.
extensionsInspect and manage extension-pack surfaces.
migrate-renamePreview or apply the supported legacy naming migration.
release-planGenerate and validate release planning data.
mark-mergedRecord a hand-merged file as permanently project-managed.

VALIDATION AND OPERATIONS

FamilyPurpose
sdlcValidate configuration, issues, paths, role passes, evidence, lifecycle events, collaboration, PRs, releases, skills, agents, audit, and migration state.
cockpitRun or diagnose the optional local Goal Command Center.
skillsSynchronize and inspect managed skill content.
pluginsBuild, validate, and inspect native harness plugin artifacts.
settingsPreview structural settings merge and validate its result.

The maintained release gate also covers executable agent evals, multi-agent acceptance, sandbox behavior, Cockpit smoke paths, publishable payload validation, native manifests, and settings merge checks. Lifecycle evidence can derive outcome measures such as cycle time, review latency, release lead time, follow-up age, and incident-to-regression-eval time.

COMMON OPTIONS

--target <path>Repository AgentFlow should inspect or manage.
--jsonMachine-readable output where the selected command supports it.
--dry-runPreview supported mutations before applying them.

Command-specific options are authoritative in node bin/cli.mjs --help for the pinned source version. Do not probe init --help or sync --help against an important project: those command parsers may execute the command rather than show nested help.

FILES

PathRole
AGENTS.mdCanonical repository policy.
agent-workflow.config.jsonProject-local workflow, branch, validation, routing, and bounded-work configuration.
agent-framework-lock.jsonManaged-file ownership and update baseline.
.agent-runs/Local scratch evidence; durable summaries belong in issue and PR records.

SAFETY AND APPROVALS

  • Inspect existing instructions and environment before installation.
  • Preserve project-owned conventions and divergent files.
  • Use one accountable executor by default; add agents only when they improve a decision.
  • Keep workflow profile separate from action authority.
  • Keep high-assurance review and release approval human-controlled.
  • Record follow-ups instead of hiding deferred work in TODOs or chat.

EXAMPLES

node bin/cli.mjs doctor-env --target C:/work/project --json
node bin/cli.mjs onboarding-prompt --target C:/work/project
node bin/cli.mjs doctor --target C:/work/project
node bin/cli.mjs update-prompt --target C:/work/project

v1.0.0 documentation ↗ · Evaluation guide · Product overview