fastbuild-agent

The /fastbuild agent. It keeps a coding agent — Claude, Cursor, Gemini, Copilot — working through a dependency-ordered task graph held by a FastBuild server, and gives it AST-backed project memory so it does not have to rediscover the codebase every session.

pip install fastbuild-agent

Use

export FASTBUILD_URL=http://localhost:8080

fastbuild adopt                      # register this repo and ingest it into memory
fastbuild backfill BACKLOG.tsv       # seed the forecast from git history
fastbuild watch BACKLOG.tsv          # have the server keep it in sync
fastbuild requirement docs/feature.md  # derive a task graph from a requirement
fastbuild next                       # claim the next ready task
fastbuild done <task-id> --changed src/a.rs
fastbuild status                     # is it moving, and if not, why

adopt works the same on a fifteen-year-old monolith as on an empty directory. Brownfield is not a mode — the ingest simply finds less in an empty tree.

The /fastbuild slash command

plugin/claude-code/fastbuild.md is the slash command definition. Copy it to ~/.claude/commands/fastbuild.md for Claude Code, or paste its body into whatever your agent calls a custom command. fastbuild prompt prints the same instructions, so every host is driven by identical text.

Why a task graph

A coding agent handed a long requirement will drift, redo work, and stall on something it cannot finish. FastBuild holds the plan instead: every task has explicit prerequisites, readiness is computed from the graph rather than stored, and a task that cannot proceed is reported with the reason. The agent's job shrinks to "do the one task I was handed", which is the thing it is good at.