Skip to content

Future Work & Ideas

This document captures ideas, potential features, and improvements to explore in future sessions. Ordered by feasibility × impact.

Priority

Codegraph Integration

Bring the codegraph pre-indexed code knowledge graph into the multi-agent loop. The Heavy-Lifter agent is the natural home — it already owns the "deep code investigation" workflow.

  • @agent-heavy-lifter:impact @symbol Foo — triggers a codegraph impact analysis (downstream callers, tests, types) and writes results to an impact-reports collection synced to the vault.
  • Codegraph as tools — expose codegraph_lookup and codegraph_callers as wiki_* tools so the Miner can resolve symbol references when extracting entities.
  • Vault sync — render codegraph findings as Obsidian Canvas (wiki_sync(format="canvas")) for visual blast-radius views.
  • Cross-project symbol linking — surface "also defined in project Y" links in entity pages.

Status: not yet started. Priority because the Heavy-Lifter's current external-delegation workflow is slow and lossy compared to a pre-indexed graph query.

Advanced Deduping

Expand dedupField logic in autopilot mode to use fuzzy matching or semantic similarity rather than strict exact string matches.

Background Auto-indexing

A watcher or background task that automatically re-indexes when files in configured collection directories change, instead of requiring manual /wiki reindex.

Pagination for query_wiki

Add limit and offset parameters. As collections grow large, returning the entire dataset bloats the LLM context.

Medium

Interactive TUI Rendering

Leverage pi's CustomMessage and rendering capabilities to display collection entries as rich tables or cards in the terminal UI, rather than raw JSON.

LLM Schema Generation (discover_schema)

A tool that reads an un-configured .jsonl file, infers the schema, and updates pi-vault-mind.config.json to map it.

pi-context Enhancements

  • Direct context_tag in append_wiki — auto-bookmark after appending.
  • context_checkout integration — switch to relevant collection state on checkout.
  • enhanceInjectors implementation — allow pi-context's active branch/tags to refine regex-based injectors.
  • context_log visualization — include collection entry summaries in log output.

Collection Migrations

A /wiki migrate command to transform entries from an old schema to a new schema when configuration changes.

Redaction / Tombstoning

A tombstone_entry tool that marks entries as deleted without breaking the append-only log format. Needed for real-world usage (accidental API key leaks, etc.).

Nice to Have

Other Extension Integrations

  • pi-web-fetch — collection entries store URLs for dynamic fetching and summarization.
  • pi-github — auto-link collection entries to GitHub issues, PRs, or code snippets.
  • Custom Linting/Validation — use collection entries as source of truth for code analysis rules.

A short .mp4 or .png demonstrating the workflow for the pi package marketplace (pi.video / pi.image in package.json).