Imported from jhnguyy/pi-env (.agents/skills/index-generator/SKILL.md). Install upstream with npx skills add jhnguyy/pi-env --skill index-generator. Copyright stays with the author.

Index Generator

Use a compressed index when passive navigation helps an agent find relevant sources without loading the full collection.

What a Good Index Is

A navigational map, not documentation. One line per entry with the most distinctive, retrievable facts — not a description of what the file is.

Format

Adapt to context. For markdown/notes vaults:

## <Group>

| Entry | [Status] | Summary |
|---|---|---|
| [[path/to/note]] | active | One-liner: key facts, numbers, decisions made |

For code or docs (pipe-delimited, AGENTS.md-style):

[<Domain> Index]
|<subfolder>:{file1 — one-liner, file2 — one-liner}
|<subfolder>:{file3 — one-liner}

Rules regardless of format:

• One line per entry, no sub-bullets
• Concrete details: names, numbers, statuses, decisions — not summaries of topics
• Status column only where entries have meaningful state (projects, tasks)
• Group by subdomain when the set is large. Link to sub-indexes rather than flattening everything.

Process

1. Enumerate — list all files/notes in scope
2. Read — read each entry that you do not know. Read any existing index first to preserve its intentional structure.
3. Write — one-liner per entry, lead with the most distinguishing fact
4. Link up — if this is a sub-index, verify the parent index references it

Maintenance

• Read before overwriting — avoids dropping entries added since last generation
• Status fields drift. Verify them from the source, not from memory.
• Remove entries for files that no longer exist
• When a new subdomain appears, create its index and add it to the parent