Repository Inventory

This is a point-in-time inventory of the versioned BaseCoat source tree. It complements the generated asset manifest, which is the authoritative file-level distribution inventory.

Snapshot

Generated from the source tree on 2026-09-01.

Asset type	Count	Source
Agents	130	agents/*.agent.md
Skills	134	skills/*/SKILL.md
Instructions	91	instructions/*.instructions.md
Prompts	6	prompts/*.prompt.md
GitHub Actions workflows	90	.github/workflows/*.yml
Documentation pages	416	docs/**/*.md before this snapshot was added
Agent evaluation definitions	229	agents/*.eval.yaml
Skill evaluation definitions	134	skills/**/eval.yaml

The docs homepage has the current distributable asset totals and is checked by scripts/validate-basecoat.ps1. This page provides additional operational counts that do not belong in the product overview.

Model Assignment

Agent frontmatter currently resolves to four canonical models:

Model	Assigned agents	Share
claude-sonnet-5	79	60.8%
gpt-5.3-codex	26	20.0%
gpt-5.4-mini	23	17.7%
gpt-5.4	2	1.5%
Total	130	100.0%

The capability catalog records 28 published models, of which 26 support Copilot CLI, 7 are CLI auto-selectable, and 15 support configurable reasoning. Runtime entitlement is user- and organization-specific; the catalog is not an allowlist.

Refresh assignment data with:

pwsh -NoProfile -File scripts/generate-model-inventory.ps1

See Model Inventory and Model Capability Framework for the generated assignment list and capability details.

Token Footprint

The validator estimates tokens as round(word_count * 1.7). This is a sizing heuristic for always-loaded entrypoints, not provider billing usage or runtime context consumption.

Asset type	Files	Words	Approximate tokens	Average tokens	At/above 630 tokens
Agent entrypoints	130	39,444	67,055	516	0
Skill entrypoints	134	31,128	52,918	395	0
Instructions	91	52,779	89,724	986	59
Prompts	6	2,185	3,714	619	2
Total	361	125,536	213,411	591	61

The validator's 630-token per-file warning applies only to agent and skill entrypoints. The instruction and prompt values in the last column are raw comparisons for visibility, not validator warnings or violations. Their loading scopes also differ, so they are not equivalent to a single session context size.

For session-level token economics and operating thresholds, see Token Optimization. Recalculate the table after material asset changes with the same -split '\s+' word-count method used by scripts/validate-basecoat.ps1.

Refresh Expectations

1. Regenerate model-map.json and model-inventory.md whenever agent model frontmatter changes.
2. Update this snapshot when an asset count, workflow count, token budget, or capability-catalog count changes materially.
3. Regenerate asset-manifest.json when tracked distributable assets change.