How It Works - Fellowship

Single Task — /quest

Run /quest for any non-trivial task. It walks through six phases with hard gates between each.

Phase 0: Onboard

Creates an isolated git worktree. Loads task-relevant context via /council. Checks for lembas checkpoints from previous sessions.

/council

GATE — requires approval

Phase 1: Research

Explores the codebase with subagents. Studies reference implementations. Extracts conventions via /gather-lore.

/gather-lore

GATE — requires approval

Phase 2: Plan

Enters plan mode. Creates implementation plan with file:line references. Presents plan for your approval.

plan mode

GATE — requires approval

Phase 3: Implement

TDD red-green-refactor. Writes failing test, implements, verifies. If stuck: commits partial work, documents blocker, returns to Plan.

test-driven-development

GATE — requires approval

Phase 4: Review

Convention review via /warden. Code quality verification. Runs full test suite.

/warden, verification-before-completion

GATE — requires approval

Phase 5: Complete

Creates PR. Cleans up worktree. Reports completion.

finishing-a-development-branch

Multiple Tasks — /fellowship

Run /fellowship for multiple independent tasks. Gandalf coordinates parallel agents.

Describe

You describe multiple tasks (quests and scouts)

Analyze

Gandalf analyzes them, creates task list

Spawn

Spawns quest-runner agents in isolated worktrees

Scout

Spawns scout agents for research questions

Monitor

Palantir monitors progress (at 2+ quests)

Gate

Gates surface to you for approval

Status

Say "status" anytime for a progress table

Deliver

Each quest produces a PR; scouts produce reports

Gandalf coordinator

Quest 1

worktree → phases 0-5 → PR

Quest 2

worktree → phases 0-5 → PR

Scout

investigate → validate → report

Palantir

monitors all active quests

Research — /scout

For questions that need investigation but not code changes.

Investigate

Explores codebase, reads docs, runs searches

Validate optional

Spawns fresh adversarial subagent to challenge findings

Deliver

Structured report with confidence levels

Gate Enforcement

Gates aren't just prompts — they're structurally enforced.

Hook-Blocked Tools

Plugin hooks block tool access (Edit, Write, Bash, etc.) after gate submission.

Lead Approval Required

Tools stay blocked until the lead agent approves by writing to the quest state file.

Verified Prerequisites

Running /lembas and updating task metadata are verified before gate submission.

No Self-Approval

Self-approval is structurally impossible. The agent that submits the gate cannot approve it.

Prompt-only compliance ~33%

Hook-enforced compliance ~95%+