indigokarasu

OCAS self-improvement orchestrator (Darwin's finch — adaptive evolution). Mines session JSONL files to detect corrections, breakthroughs, methodologies, course-changes, and behavioral directives (Always/Never). Routes each finding to the optimal storage tier: MEMORY.md, skill files, reference files, or Chronicle KG. Compacts MEMORY.md by routing entries to the correct tier. Part of the OCAS System Evolution Layer alongside Mentor, Fellow, and Forge. NOT for real-time behavioral adaptation, skill evaluation, or skill creation.

Structured OSINT research on people, companies, and organizations. Use for provenance-backed briefs, entity resolution across public sources, background research with cited sources, or free-first research workflows that escalate to paid sources only with explicit permission. Do not use for topic research without a person/org focus (use Sift) or illegal data collection.

Self-improving orchestration and evaluation engine. Manages long-running multi-skill workflows, analyzes journals from all skills, evaluates champion vs challenger variants, and proposes skill improvements to Forge. Use for multi-step project management, heartbeat runs, skill performance evaluation, or multi-skill coordination. NOT for: web research (use Sift), skill building (use Forge), user communication (use Dispatch), real-time skill execution, content generation, system health monitoring (use Custodian), or skill evaluation scoring.

Calendar management. Use for viewing, querying, creating, modifying, deleting, or analyzing calendar events. Handles natural-language scheduling, conflict detection with flexibility classification, free slot finding, automatic travel time event insertion between consecutive appointments, recurring event management, and daily schedule briefings for Vesper. Do not use for reminders without calendar context, task management, or general time/timezone questions.

Live world-data query engine. Queries real-time external APIs for factual ground truth — no synthesis, no opinion, no research. Routes requests through a registry of ~55 registered sources covering US government data, scholarly literature, weather and hazards, geocoding, finance and macro indicators, court records, nutrition, news events, property records, land due-diligence, academic papers, satellite imagery, and product manuals (3M+). Do not use for web research (use Sift), entity investigations (use Scout), or pattern analysis over historical signals.

Accomplishes tasks that need tools the agent does not have. Plans approaches, fills capability gaps with sandboxed tools, executes within an isolated session. Disposable identity, parallel discovery, incremental checkpoints, graceful degradation. Output: task result plus replay script. No global installs, no real identity, clean state in and out. Do not use for tasks solvable with installed skills, permanent skill installs, skill builds (use Forge), or general web research (use Sift).

Use when checking appointment availability, booking services, monitoring for openings, or discovering venues at salons, spas, and restaurants. spot.discover finds and compares venues via Yelp before booking. Supports Acuity Scheduling, Square Appointments, Resy, Tock, SevenRooms, OpenTable, Meevo, Vagaro, Mindbody, Fresha, StyleSeat, Calendly, Yelp Reservations, Booksy, GlossGenius, SimplyBook.me, Boulevard, Mangomint, DaySmart, ResDiary, and Eat App. Integrates with ocas-vpn for bot block bypass. NOT for general travel planning (use ocas-voyage), calendar sync (use ocas-sands), restaurant reservations on unsupported platforms. Trigger phrases: 'book an appointment at', 'check availability at', 'when can I get a [service]', 'find me a slot at', 'is [venue] available', 'watch [venue] for openings', 'alert me when [venue] has availability', 'monitor [venue]', 'find a restaurant in', 'compare salons near', 'discover [type] near'.

Empirical experimentation engine. Invoked by Mentor to evaluate, compare, and promote improvements to OCAS skills, prompts, heuristics, and workflows using benchmark-driven experiments. Returns best variant result with lineage. Not user-invocable — called only by Mentor.

Art-direction engine for text-to-image generation. Separates Style from Content to keep aesthetic identity consistent across an image series. Use when generating, rendering, or illustrating images (especially across a series), capturing the style of a reference image for reuse, or producing an image in a named/saved style. Not for captioning, image analysis for decisions (use ocas-look), or photo editing.

Converts user-provided images into validated, decision-ready action drafts. Routes images by inferred intent across domains: events from flyers, macros from meal photos, places to save, products to price, receipts to log, documents to file, civic issues to report. Supports reverse image search via Yandex and Google. NOT for generic OCR, computer vision research, or surveillance.

Nightly journal curator. Batch-processes OCAS skill journals via relevance classification and writes curated content to journal files for the configured memory provider to ingest. Classifies each journal for filing as a verbatim journal note, structured entity/relationship data, or skip. Features re-emergence detection, two-pass stale handling, change magnitude gates, hibernation protection, and incremental cursor-based resumption. NOT for real-time memory filing, skill evaluation, behavioral pattern detection, or entity identity resolution.

Bounded behavioral refinement loop. Records outcomes, extracts micro-lessons from repeated patterns, consolidates them into capped active behavior shifts, applies shifts at runtime, and generates plain-language debriefs. Use for recording task outcomes, extracting lessons from repeated patterns, managing active behavior shifts, generating runtime briefs, or producing debriefs. Not for: general memory (use Chronicle), preference tracking (use Taste), real-time task execution, content generation, system health monitoring (use Custodian), or skill evaluation scoring (use Mentor).