/ stack
The toolchain I actually run.
The toolchain I run day-to-day. Nothing aspirational, only what's actively in use.
Alfred
specialist agents
each one has a single job, planning, implementing, reviewing, or checking, and none reviews its own work
gated pipeline
work moves forward only after another agent checks it, not because I decided it was ready
persistent memory
decisions and lessons are recalled into every new session, not re-explained from scratch
independent QA
a separate agent re-runs the checks and decides when work is actually done
Editor
Claude Code
the substrate Alfred runs on; every session goes through it
JetBrains IDEs
IntelliJ, GoLand, PyCharm, DataGrip · surgical edits, debugging, and SQL work
Prompts I treat as code
reusable prompts
prompts and playbooks versioned and reused across agents, not a one-off per project
review and ship discipline
comes from Alfred's own gated workflow rather than a slash-command I rebuild by hand
prompts versioned like code
committed and reviewed the same way code is, not held in someone's head
Multi-agent orchestration
Opus for the hard calls
ambiguity, architecture, and anything that needs judgment go to Opus
Sonnet for execution
once work is well-specified, Sonnet implements it: cheaper, faster, safe to run in parallel
model fits the job
each role runs on the model suited to its work, so I'm not choosing one by hand every time
Supporting tools
iTerm + zsh
still where the agents live
Linear + Slack
team coordination
Obsidian
writing first drafts before they hit /writing
Where I don't use AI
production secrets
never paste credentials, even into a sandbox
live incident triage
raw logs first, agent assistance second
commit message archaeology
git blame is faster than asking