section · guides
Cookbook
Task-oriented recipes — "I want to do X." One page per job-to-be-done.
Cookbook pages start from your goal, not the tool. Pick the one that matches what you want to do.
Customize the team
- Add a custom agent — define a new specialist (registry, prompt, sync, smoke test).
- Custom specialists and teams — author your own specialist or team into
.construct/org/or~/.construct/org/, without touching the built-in roster. - Inspect running agents — see what specialists ran for a task and why.
Fix something
- Fix a policy violation — a commit, push, or PR is blocked. Find which gate fired and what to do.
- Slash command index — find the right slash command by intent.
- Operations → Troubleshooting — common failure modes beyond policy enforcement.
Connect to external systems
- Manage providers — the provider model in one page.
- Configure GitHub
- Configure Slack
- Configure Jira and Confluence
- Configure Salesforce
Plug in your own pieces
- Plug in your own LLM — swap models per tier (reasoning / standard / fast). Anthropic, OpenAI, OpenRouter, Ollama, any OpenAI-compatible.
- Plug in a retrieval backend — embedder, chunker, indexer, fuser, reranker, compressor. Six layers, all swappable.
Workflow templates
- Onboard a new engineer — day-1 brief, access checklist, week-1 handoff in one command.
- Track research findings — capture, stamp, and cite research notes so they survive into PRDs and ADRs.
- Cross-team handoff — transfer in-flight work without losing context, blockers, or decisions.
Common day-to-day tasks
- Generate artifacts (PRDs, ADRs, RFCs)
- Use the inbox — drop a file, watch the daemon classify it into R&D triage, drive the queue with
construct intake - Query the knowledge base
- Distill and infer from documents
- Wireframe and drop assets
- Observability and cost — Telemetry traces (now including R&D-loop events), cost ledger, eval datasets
- Start embed mode
- Override storage root