Skip to content

Latest commit

 

History

History
45 lines (32 loc) · 3.39 KB

requirements.md

File metadata and controls

45 lines (32 loc) · 3.39 KB

Requirements

/sdlc:setup checks this machine for all of this and reports what's missing, too old, optional, or a recommended companion. Run it once per machine after installing the plugin.

Platform

Linux only. On any other operating system, /sdlc:setup reports it as not supported yet, because the dispatch scripts need setsid, GNU stat and GNU date; the planning skills (design, split, prior-art) still work.

Required

Tool Minimum version Used for
Claude Code 2.1 runs every skill and every worker session
beads (bd), Dolt backend 1.2.2 tasks, dependencies, metadata, gates, merge-queue locks, the audit trail
worktrunk (wt) 0.77 one worktree per task, project hooks (.config/wt.toml), merge approvals
git 2.30 worktrees, branches, diffs, rebases, merges
jq 1.6 reading and writing bead JSON in the shell scripts
python3 3.9 the scripts that walk the task graph, total costs, or render logs (tasks.py, stats.py, logs.py, supervise.py, finish-task.py, workers.py)
uuidgen, setsid, pgrep, timeout — worker session ids; detaching a worker from its parent; detecting whether a worker process is still alive; bounding a verify command's runtime

Optional

Tool Needed for Without it
gh epic-pr integration mode: pushing the epic branch and opening the pull request epic-pr mode isn't usable; direct and epic-merge still work
lizard (pip install lizard) metrics.sh: complexity findings in /sdlc:review, and the branch analysis (/sdlc:analyse) those checks are skipped, with a one-line notice, instead of failing

Recommended companions

Not required; each makes the workflow better or cheaper. /sdlc:setup names each one that isn't installed and offers to help find it:

Companion Why
rtk compresses command output, so every session and worker spends fewer tokens (not reachingforthejack/rtk, an unrelated project with the same command name)
rg (ripgrep) faster text-search fallback for prior art than grep, and skips ignored files
a reviewer agent (~/.claude/agents/reviewer.md or .claude/agents/reviewer.md) a worker's --review agent runs this agent instead of a plain model call
roborev /sdlc:review runs roborev review --local on the same diff as a second opinion, often from another model, in a project with a .roborev.toml
specialist agents (any .md file under ~/.claude/agents or .claude/agents) a worker hands implementation to a matching agent when the project's CLAUDE.md asks it to
a project debugging skill named with bd config set custom.dispatch.debug "use the <skill> skill", so a worker whose verify command fails is told which one to reach for

Version note

sdlc is developed against Claude Code 2.1.272. /sdlc:setup's version checks (claude --version, bd version, wt --version, git --version, jq --version, python3 --version) enforce only the minimums in the table above.

See settings for how to configure what's installed, and develop the plugin for what a contributor additionally needs.