afc

name	afc
description	Use when they want to quickly load a repository's documentation and source code into the context for implementation-detail questions. Runs scripts/afc.py to emit a budgeted one-shot context dump, defaulting to 160k tokens.

Aggressive Fill Context

Purpose

Use this skill when the user wants fast repository context for questions, design checks, or implementation-detail confirmation. The goal is to avoid slow file-by-file exploration by dumping as much relevant markdown and source code as the context budget allows.

Quick Start

Resolve scripts/afc.py relative to this SKILL.md, then run it from the target repository:

python3 /path/to/afc/scripts/afc.py

Defaults:

• Budget: 160k tokens, converted to 640000 bytes.
• Preference: both, meaning documentation and source are both eligible.
• Root: current working directory.

Common runs:

python3 /path/to/afc/scripts/afc.py 200k
python3 /path/to/afc/scripts/afc.py --bytes 800k --prefer source
python3 /path/to/afc/scripts/afc.py --tokens 120k --prefer docs --root /path/to/repo

Bare budget values such as 200k mean tokens. Byte budgets must use --bytes or a byte suffix such as 640kb.

Workflow

1. Identify the target repository. If the user did not specify one, use the current workspace.
2. Choose the budget. Use the user's context-window size when provided; otherwise use the default 160k tokens.
3. Choose --prefer docs, --prefer source, or --prefer both. If the combined docs and source output fits, AFC includes both even when a preference is set. If the combined output does not fit and a single category is preferred, AFC only selects from that preferred category.
4. Run the script and use its stdout as the loaded context. Set the terminal output allowance high enough for the requested budget when your environment requires it.
5. Answer the user's question from the loaded context. Only inspect individual files manually if AFC reports no candidates, the budget is too small, or the user asks for targeted follow-up work.

Script Behavior

scripts/afc.py collects repository files in two groups:

• Docs: markdown-style files such as .md, .mdx, .markdown, and standard root docs such as README.md and AGENTS.md.
• Source: common programming, config, schema, shell, web, and build files.

The script excludes dependency directories, build outputs, caches, lockfiles, minified bundles, source maps, binary files, and non-UTF-8 files. It respects Git's tracked, untracked, and .gitignore-excluded file view when the root is inside a Git worktree; otherwise it uses a deterministic filesystem walk with the same exclusion rules.

Output is a markdown context dump with:

• Budget and discovery metadata.
• Included file inventory.
• Omitted counts.
• Full contents for each included file, wrapped in safe code fences.

The script enforces the byte budget on its final stdout. If the preferred category alone is too large, it greedily keeps the highest-priority complete files that fit; it does not emit partial source files.