How Aspens Saves Tokens

Without context, coding agents burn through tokens in three ways:

1. Searching — Bash/Grep calls to find relevant files, drilling through directories, reading file after file trying to understand the codebase structure
2. Reading — opening files that turn out to be irrelevant, reading entire files when only the function signature mattered
3. Rebuilding — writing new components from scratch because it didn’t know an existing one already does the same thing, duplicating utilities, creating inconsistent patterns

On a large codebase, a significant portion of tool calls are pure exploration — not writing code.

How aspens fixes this

Skills eliminate searching

When Claude has a billing skill that says “Stripe wrapper is at src/services/billing/stripe.ts, usage checker is at src/services/billing/usage.ts,” it doesn’t need to Grep the codebase looking for Stripe-related code. It already knows.

The import graph prioritizes what to read

Instead of randomly opening files, the graph tells the agent which files are architecturally important (high fan-in = many dependents). It reads the files that matter instead of skimming many.

Critical rules prevent rebuilding

“Don’t call Stripe API directly — always use the wrapper in stripe.ts” prevents the agent from writing a new Stripe integration when one already exists. “Usage counters use cache first, DB fallback” prevents a direct DB query.

Scoped context = less context loaded

Skills auto-trigger based on file patterns. When editing billing code, only the billing skill (~35 lines) loads. The auth skill, the courses skill, and the onboarding skill stay out. Less context loaded = fewer tokens consumed per prompt.

The effect

Without aspens, sessions start with exploration — search calls to find files, reads to understand patterns, and occasional rewrites when the agent discovers it duplicated something.

With aspens, sessions start informed — skills point directly to key files, graph ranking guides what to read first, and anti-patterns sections prevent common mistakes.

The token savings compound across sessions. Every session without aspens starts from zero. Every session with aspens starts informed.

Doc sync keeps it current

The savings only work if skills are accurate. Stale skills are worse than no skills — they point at wrong files and outdated patterns.

aspens doc sync runs after every commit (via git hook), reads the diff, and updates affected skills automatically.