sjawhar

Use when the user wants to turn research papers into a podcast episode (or "make a podcast" / "paper podcast" / "audio digest" / "research digest"). Works for any field. Takes a user-supplied paper source (arXiv query, arXiv IDs, or a PDF list), confirms the lineup with the user, extracts a faithful ground-truth JSON, grounds each paper against related work, writes a long-form two-host script, and renders an MP3 with Gemini 3.1 Flash TTS.

Use when reading, searching, or updating Airtable records — applicant/hiring data or any Airtable base. Triggers on airtable.com URLs, base IDs (app...), table IDs (tbl...), or requests to list, query, filter, or update Airtable records.

Use when performing ANY version control operation, starting a work session, checking repo state, or orienting to a codebase. This user uses jj instead of git — NEVER use git commands. Triggers on: commit, push, pull, branch, checkout, rebase, merge, diff, log, status, stash, reset, cherry-pick, bookmark, workspace, conflict resolution, 'what's the repo state', 'are other agents working here', 'what branches exist', 'starting work', 'orient me'.

Use when reading Slack messages, searching conversations, sending messages, listing channels, or interacting with Slack workspaces

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

Use when searching flights, hotels, or rental cars; comparing fares across flexible dates; discovering cheap destinations from a fixed origin; or hunting hidden-city ticketing deals. Trigger on multi-city itineraries, fare calendars, "where can I fly cheaply", price-sensitive trip planning, or any time the user wants a sanity-check against Google Flights pricing — Skiplagged surfaces hidden-city deals other engines deliberately hide.

Search the web via Ceramic Search (lexical/keyword-based). Use when looking up current events, recent news, time-sensitive facts, specific people/products/companies, technical docs, or any topic requiring fresh web results. Triggers on "search the web", "look up", "find recent", "latest news", "current", or when built-in knowledge is likely stale.

Process ghost-wispr summary-ready Envoy events into GitHub issues and Slack summaries. Use when setting up or running a persistent OpenCode session subscribed to notifications.ghost-wispr.summary-ready.

Use when subscribing sessions to Envoy topics, sending agent-to-agent messages, or reasoning about topic formats for Slack/GitHub/agent routing.

Use when coordinating Legion workers across issues, dispatching workers, monitoring progress, or routing triage items

Use when dispatched by Legion controller to work on an issue in a jj workspace

OpenCode serve HTTP API reference. Use when interacting with sessions, reading messages/todos, aborting workers, managing workspaces, or any direct OpenCode serve API call. Each `opencode serve` process is independent with its own port; Legion runs one shared serve per daemon (default port 13381, may differ if multiple daemons coexist on the host).

Capture learnings from completed work via dual-perspective retrospective. Invoked by resuming an implement worker session — the implementer has full context, and a fresh subagent provides an outside view.

Manage GitHub issues via Projects V2. Use when LEGION_ISSUE_BACKEND=github.

Research institutional knowledge before escalating questions to users. Check docs/solutions/ and codebase patterns before asking humans.

Manage Linear issues. Use when working with tasks, tickets, bugs, or Linear.

Use when reading or writing MacroFactor nutrition, food logs, weight, workout, or training program data. Triggers on meal tracking, macro counting, calorie logging, body weight, exercise logging, food search, or any MacroFactor interaction.

Use when someone asks anything about bench press form or performance: bench technique, setup, grip width, bar path, arch, leg drive, shoulder position, elbow tuck/flare, touch point, pause vs touch-and-go, incline/decline/reverse-grip choices, shoulder or elbow pain while benching, fixing sticking points, bench accessories, or programming to bench more weight.

Use when someone asks anything about deadlift form, deadlift technique, conventional vs sumo deadlift, setup, hip hinge, bar path, back rounding, lockout, deadlift grip, deadlift programming, fixing deadlift sticking points, Romanian deadlift, stiff-leg deadlift, deadlift accessories, or how to pull more weight.

Evidence-based nutrition and dieting guidance for body composition and performance. Use this whenever users ask about nutrition, calories, macros, protein targets, fat loss, cutting, bulking, recomping, meal timing, intermittent fasting, keto/low-carb, processed foods, diet adherence, plateaus, metabolic adaptation, weight-change pacing, or food-choice tradeoffs. Trigger aggressively for questions about what/how much to eat, how to lose fat or gain muscle, or how to adjust diet plans over time—even when users do not explicitly ask for "diet coaching."

Use when anyone asks about squatting in any capacity: squat form or technique, setup, bar position (high bar vs low bar vs front squat), stance width, toe angle, depth, knee tracking, back angle, bracing, ascent/sticking point cues, fixing squat problems (knee cave, butt wink, forward lean, balance loss), mobility limits, accessory selection, or squat programming for powerlifting, weightlifting, bodybuilding, or general strength.

Use when someone asks how to get stronger with evidence-based programming: exercise selection, progressive overload, set/rep schemes, training volume, frequency, intensity (%1RM/RIR), hypertrophy planning, periodization, deload/fatigue management, and beginner/intermediate/advanced progression. Trigger this for workout-structure questions like "how many sets," "how often should I train," "what reps/intensity should I use," "how do I break plateaus," or "how should I set up a full program."

Use when posting a daily standup, status update, or YTH report to Slack using time-tracker activity data

Use when analyzing tt classify output to identify work streams, assign events to streams, and generate time reports. Triggers on weekly review, standup prep, or explicit stream inference requests.

Use when the user asks to "run weekly review" OR wants to reflect on the past week with historical trend analysis and time tracking data. Trigger phrases: "weekly review", "review the week", "weekly reflection", "how did last week go".

Use when the user asks what should I do, what's next, todos, priorities, priority alignment, task ranking, deferred work, or am I drifting.

Use when writing Pivot pipeline stages, seeing annotation errors (Dep, Out, Annotated), loader mismatches, "cannot pickle" errors, DirectoryOut validation failures, or IncrementalOut path mismatches

Use when testing Pivot fingerprint correctness against a real pipeline, hunting for undetected code changes that cause stale cached outputs, or validating Change Detection Matrix claims against production code