How Charter Works

Charter is the governance layer that turns your project's policy document (charter.md) into structured context that every agent mission action automatically receives. This page explains the mental model. For a step-by-step walkthrough, see the Governed Charter Workflow Tutorial.

Key invariant: .kittify/charter/charter.md is the human policy source of truth. Do not hand-edit derived files such as governance.yaml, directives.yaml, metadata.yaml, references.yaml, context-state.json, synthesis manifests, or provenance sidecars. Agent synthesis input under .kittify/charter/generated/ is produced by the harness, not by routine operator edits.

What Charter Does

Charter solves a specific problem: agent prompts need consistent, project-accurate policy context, but that context must come from a single authoritative source that you as an operator control.

The mechanism:

You write (or generate via interview) a charter.md file that captures your project's policy decisions — testing standards, quality gates, branching rules, directive selections.
The synthesis pipeline reads that file and produces a structured charter bundle of YAML artifacts in .kittify/charter/.
When spec-kitty next invokes an agent profile for a mission action, the runtime injects the relevant charter context into the prompt automatically. The agent does not invent governance; it reads and complies with what the charter says.

Synthesis Flow

The full Charter setup flow uses these commands in sequence:

# Step 1 — Capture policy decisions interactively (or use --defaults for CI)
uv run spec-kitty charter interview

# Step 2 — Generate charter.md and initial bundle from interview answers
uv run spec-kitty charter generate --from-interview

# Step 3 — Check for graph-native decay (orphaned directives, contradictions, etc.)
uv run spec-kitty charter lint

# Step 4 — Validate + promote agent-generated doctrine artifacts to .kittify/doctrine/
uv run spec-kitty charter synthesize

# Step 5 — Validate the charter bundle against the CharterBundleManifest v1.0.0 schema
uv run spec-kitty charter bundle validate

# Check sync status at any time
uv run spec-kitty charter status

charter context is a separate runtime/debug command for rendering action-specific governance context for a specific workflow action. It is not part of the synthesis pipeline:

# Render what governance context an agent would receive for the 'implement' action
uv run spec-kitty charter context --action implement --json

After editing charter.md by hand, re-sync the YAML config files with:

uv run spec-kitty charter sync

For partial regeneration of a specific directive or tactic without touching unrelated artifacts:

uv run spec-kitty charter resynthesize --topic directive:PROJECT_001

The DRG-Backed Context Model

The charter bundle is not a flat file — it is backed by a Directive Relationship Graph (DRG). The DRG is a directed graph whose nodes are directives, tactics, and glossary terms, and whose edges use typed relations such as scope, requires, suggests, vocabulary, instantiates, replaces, and delegates_to.

When spec-kitty next prepares a prompt for a mission action (for example, implement), the runtime traverses the DRG from the entry point for that action and collects the relevant subgraph. This is governed profile invocation: the agent receives a (profile, action, governance-context) triple, where the governance context is the DRG-derived subgraph rendered as structured text.

The agent cannot see or modify the DRG directly. It receives the rendered context and acts in accordance with the directives it finds there.

Bootstrap vs Compact Context

The DRG traversal for a given action can produce large context payloads for complex projects.

Bootstrap mode: the first time an action loads context (or when the charter is freshly synthesized), the runtime injects the full relevant DRG subgraph. The agent sees all applicable directives and tactics.
Compact-context mode: when the DRG context payload is too large to include in full, the runtime falls back to a summarized view — resolved paradigms, directives, and tool list only, without the full library text. This is a known limitation (see issue #787 in the project tracker; check that issue for current resolution status).

Do not assume full-context behavior unconditionally. Large governance layers may trigger compact-context mode, causing agents to receive less detail.

Key Governance Files

File	Written by	Purpose
`.kittify/charter/charter.md`	Human	Policy source of truth — the only file you edit
`.kittify/charter/governance.yaml`	Auto-generated (`charter sync`)	Testing, quality, performance, branch, and doctrine-selection config
`.kittify/charter/directives.yaml`	Auto-generated (`charter sync`)	Extracted directive list
`.kittify/charter/metadata.yaml`	Auto-generated (`charter sync`)	Charter hash, extraction timestamp, and parser metadata
`.kittify/charter/references.yaml`	Auto-generated (`charter generate`)	Reference manifest for shipped and local doctrine content
`.kittify/charter/generated/`	Agent harness	Candidate doctrine YAML consumed by `charter synthesize`
`.kittify/charter/synthesis-manifest.yaml`	Auto-generated (`charter synthesize`)	Manifest for promoted project-local doctrine artifacts
`.kittify/charter/provenance/*.yaml`	Auto-generated (`charter synthesize`)	Provenance sidecars for synthesized doctrine artifacts
`.kittify/doctrine/`	Auto-generated (synthesize)	Project-local doctrine promoted by synthesizer

See Governance Files Reference for the full table.