/edikt:context

Loads your project's memory into the current session.

When to use it

New session, existing project. A new session carries nothing forward from the previous conversation. What survives is the compact auto-memory snapshot — project name, stack, active plan, hard invariants. Run /edikt:context (or just say "remind yourself") to load the full set on top of it: product spec, active PRDs, current plan phase, every decision and its reasoning, and which rule packs are live.

After context compaction. Long sessions hit the model's context limit and get compacted. Run context to reload what matters.

Onboarding a teammate. They open the project, run /edikt:context, and the model is immediately useful — no hand-holding, no re-explaining the architecture.

What it loads

Source	What the model learns
docs/project-context.md	Project identity, stack, non-negotiables
docs/product/spec.md	What you're building and why
docs/product/prds/	Active feature requirements
docs/product/plans/	Current plan + phase progress
docs/architecture/decisions/	Architecture decisions and their reasoning
docs/architecture/invariants/	Hard constraints — non-negotiables the model must never violate
.edikt/config.yaml	Ticket system config (Linear/Jira/GitHub)
.claude/rules/	Which packs are active; flags manually edited files

What it looks like

You: remind yourself about this project

Claude: (runs /edikt:context)

Loaded context for Orders API:

  Project:   Go REST API for order management
             DDD with bounded contexts (orders, inventory, billing)
             Chi router · PostgreSQL · Hexagonal architecture

  Active plan: PLAN-bulk-orders
  Progress:    Phase 3 of 4 in progress (HTTP handler)

  Rules:     code-quality · testing · security
             error-handling · go · chi

  Decisions: 3 ADRs on file
             — hexagonal architecture
             — error wrapping strategy
             — JWT auth pattern

  Product:   spec + 2 active PRDs

Ready. What are we working on?

Auto-memory

After running /edikt:context, edikt writes a compact snapshot to the model's auto-memory (~/.claude/projects/.../memory/MEMORY.md). This file is automatically loaded at the start of every future session — so the model knows the project name, stack, active plan, and hard invariants without you running /edikt:context first.

The SessionStart hook checks if memory is stale (>7 days old) and prompts you to refresh. The memory file is local to your machine, not committed to git.

You usually don't need to run this manually

The SessionStart hook installed by /edikt:init fires when you open the project. If auto-memory exists and is fresh, it confirms context is loaded. If memory is missing or stale, it prompts you to run /edikt:context. Running it explicitly gives you the full visible output — useful when onboarding a teammate or starting a complex session.

Arguments

Argument	Description
(none)	Load full context (default)
--depth=full	Everything: project context, all decisions, all invariants, product, PRDs, plans, rules
--depth=focused	Project context, current plan phase, relevant decisions, all invariants, rule names
--depth=minimal	Project context, current plan phase title + tasks, all invariants only

--depth=full is the default. On large projects (>15 ADRs or >5 PRDs), edikt suggests --depth=focused before proceeding.

Focused mode (v0.6.0)

--depth=focused is now plan-scoped: it scans the active plan's current phase for PRD-NNN and SPEC-NNN identifiers and loads only the referenced sidecars.

The active plan is the signal for what's being worked on right now. If your phase prompt says "implements PRD-005's FR-002 and FR-003," only PRD-005's sidecar gets loaded — not every PRD in the project. Same for SPECs.

PRDs (plan-scoped — active phase references):
  PRD-005 (Webhook delivery) — 5 FRs, 8 ACs — accepted
  PRD-007 (Retry budget)     — 3 FRs, 5 ACs — accepted

If no PRD or SPEC identifiers appear in the current phase, focused mode falls back to listing all PRD titles without loading their content. full and minimal modes are unchanged.

This keeps the context budget lean on projects with dozens of PRDs while still surfacing exactly what the current phase needs.

Natural language triggers

• "remind yourself"
• "load context"
• "what's this project?"
• "catch me up"