FIL Coding Agent — AI skill for Claude Code

FIL Coding Agent Edition · V1.4.0-beta

"Governance over context." Adapted from FIL Framework V3.4.2 for long-running coding agent workflows.

---

What This Is

A governance framework for coding agents on long-running projects. Not another context dump. Not a prompt template.

**The core insight:** as a codebase grows, the problem isn't that the agent lacks context. It's that all context looks equally important. FIL fixes that by making explicit what governs decisions vs. what merely informs them.

---

The Seven Files

File	Role	Authority	Changes
CLAUDE.md	Governance layer — architectural law	L1 within repository governance	Rarely
AGENTS.md	Operational procedures — session workflow	L2	Occasionally
DECISIONS.md	Architectural decisions — permanent ADR record	L2	On new decision
DYNAMIC.md	Operational memory — current state	L5	Every session
skills.md	Domain knowledge — informs decisions	L4	On update
LOG_ERRORS.md	Prevention knowledge — distilled lessons	Prevention layer	As errors occur
BOOT.md	Initialization wizard — setup only · generates all files	—	Rarely

---

How It Works — Full Flow

1. Setup (once)

Copy `BOOT.md` to your repo root, then tell your agent:

Load BOOT.md and follow the FIL Boot sequence.

The agent interviews you one question at a time — about 18 questions — and generates all 6 project files fully populated, including the security-by-design baseline. No `[placeholder]` to fill manually. It then proposes a first commit: `chore: initialize FIL Coding Agent V1.4.0-beta`. You review and confirm.

→ See [**Setup**](#setup) below for full details (Option A guided · Option B manual).

---

2. Every working session

At startup, the agent automatically executes Step 0:

① Loads CLAUDE.md — reads your architectural principles, constraints, coding standards
② Loads DECISIONS.md — applies permane