Skip to content

interface.md

Latest commit

 

History

History
72 lines (53 loc) · 1.89 KB

interface.md

File metadata and controls

72 lines (53 loc) · 1.89 KB

Interface

Usage examples from Telegram.

Chat

> research the top 5 competitors in the AI code editor space and summarize their pricing
  [searches web, reads pricing pages, compiles data]
  Here's the breakdown: Cursor ($20/mo), Windsurf ($15/mo)...

> write a twitter thread about why developers should use AI agents
  [drafts thread with hook, insights, CTA]
  Done. 8 tweets. Want me to adjust the tone?

> deploy the staging branch and run the migration
  [pulls branch, runs deploy script, executes migration]
  Deployed. Migration applied — 3 tables updated, 0 errors.

Evolve

> /evolve
  [scans GitHub trending, HN — scores 12 signals]
  2 new skills built and auto-installed: api-rate-limiting, structured-logging

Absorb

> /absorb https://github.com/trending-project
  [clones repo, scans architecture, diffs patterns]
  Implemented 3 improvements: retry logic, health checks, graceful shutdown.

Learn

> /learn https://github.com/some-saas-boilerplate
  [deep-dives codebase, extracts patterns]
  ADOPT: their auth flow. STEAL: the webhook retry pattern. SKIP: their ORM choice.

Recall

> /recall rate limiting
  Recall: rate limiting
  5 results across 3 layers

  ## Past Conversations
  - discussed nginx rate limiting config (session: API hardening, 2026-03-10)

  ## Absorbed Knowledge
  - token bucket with sliding window, 429 retry-after header (from: github.com/foo/bar, verdict: STEAL)

  ## Observed Patterns
  - prefer per-route rate limits over global (conf: 0.7, seen 3x)

Reply Context

Reply to any bot message to use it as context for your next command:

[bot message about a repo with URL]
  -> reply: /absorb       (absorbs the repo from the replied message)
  -> reply: /learn         (learns from it)
  -> reply: /notify 30m   (reminds you about it in 30 min)
  -> reply: tell me more   (Claude sees the replied message as context)