Naming the Machine

Invocation

naming-the-machine [infrastructure|projects|workflows]

Default: full audit. Unrecognized scopes fall back to full audit.

Signal Levels

Execution

Scan

Start from what exists, skip what doesn't, never error on missing paths.

Sources (check all that exist; scope narrows which to prioritize):

• Always: CLAUDE.md, skills index, context files, memory files
• Projects: Git repos or dirs with CLAUDE.md/package.json -- prefer diversity over recency, cap at 20, read READMEs/architecture docs only
• Infrastructure: Cron jobs (crontab -l), systemd services, MCP configs, tool registries
• Workflows: Slash command tables, agent dispatch patterns, session save/load patterns

Use subagents for parallel scanning. Read structure first; drill into files only when they show signals. First run: scan broadly. Subsequent runs: focus on changes since last run.

Cross-scope stitching: After scanning individual layers, look for patterns that span layers (e.g., a cron -> agent -> skill chain). Composite patterns are often the most valuable and least visible.

Disqualify

A pattern must pass all three gates:

1. 3+ distinct components — single tools or configs don't qualify; aliases for the same thing don't count as separate components
2. Repeating — used across multiple sessions, projects, or time periods (within-session recurrence alone is insufficient)
3. Referenced or reused — the user has invoked, described, or built on this pattern (not just theoretically valuable)

If it involves client work, annotate "verify IP ownership" but don't disqualify.

Name

For each surviving pattern:

## [Working Title]
**Signal:** named | emergent | nascent
**Value lever:** One sentence — what outcome does this arrangement produce?
**Next action:** Ship it / name it publicly / add [specific missing piece] / watch it

For named items, add: Linchpin — the one component that, if pulled, the whole arrangement falls apart.

Naming guidelines: Name the value, not the mechanism. Two words preferred, three max. It should make someone ask "what's that?"

Summary Matrix

| # | Working Title | Signal | Value Lever | Next Action |
|---|--------------|--------|-------------|-------------|

Sort: named → emergent → nascent.

Persist and Diff

Save to <your-output-dir>/naming-YYYY-MM-DD.md. On first run, just save.

On subsequent runs, find the most recent prior file and diff:

• New — not in prior run
• Promoted — signal level increased
• Retired — in prior run, no longer present

Include delta in output.

Output

NAMING COMPLETE: [count] expressions identified
named=[n] emergent=[n] nascent=[n]
DELTA: [n] new, [n] promoted, [n] retired (omit on first run)