Commands¶

A command is the slash-command layer of CRASH — what you actually type in an AI IDE (/g-<name> in Claude Code, @g-<name> in Cursor and most of the rest of the supported platform roster) to dispatch a piece of gald3r's workflow. Each command is a thin dispatcher onto a skill; typing /g-status and letting your agent's own judgment invoke the g-skl-status skill unprompted produce the same result. Commands exist so you have a short, memorable name for the workflows you reach for most.

This page is the narrative layer. For the exhaustive, generated list of every command this repo ships — full descriptions, arguments, and per-platform invocation syntax — see the Commands reference. This page is about what commands are and when to reach for one; that page is the inventory.

When to use a command vs. letting your agent decide¶

Most of the time you don't need to think about this — describe what you want, and your agent's own judgment matches the right skill's description and runs it. Reach for an explicit command when you want to force a specific workflow rather than let judgment pick one, or when you want the shorter, muscle-memory invocation instead of typing out a full request. Both paths end up running the same skill.

Command families worth knowing on day one¶

Task and bug work — the everyday loop:

/g-task-add — create a new task (ID, complexity score, file, index entry) in one step.
/g-bug-add — report a bug the moment you find it, so it never gets lost in a chat transcript.
/g-status — see where the project actually stands: task/bug counts, what's blocked, what's waiting on you.

Implement/review pipelines — the g-go family, covered in depth on Autopilot:

/g-go — implement the next eligible task or bug, then auto-spawn a fresh, independent reviewer.
/g-go-code — implement only, through to awaiting-verification; never runs review itself.
/g-go-review — verify tasks/bugs awaiting review in a fresh session, with no self-review.
/g-go-go — the autopilot loop: implement and review in parallel until a hard stop.

The decision toolkit — five commands, each a real structured reasoning framework rather than a vague "think about it" prompt:

/g-thinker — lay out assumptions, weigh two angles, stress-test the first instinct, then land a clear verdict.
/g-ooda — Observe raw facts, Orient on context and bias, Decide with reasoning, Act, plus the signal to re-loop.
/g-mdmp — a seven-step staff-planning process (mission analysis through orders production) for larger or more ambiguous efforts, handing off into task creation.
/g-ipb — a four-step landscape/competitive-analysis process that ranks likely and most-dangerous outcomes, feeding into /g-ooda or /g-mdmp.
/g-stz — apply classical strategy doctrines to the situation at hand, naming the principle behind each recommendation, with a guard against any real-world manipulation reading.

Session control — keeping a long session honest about what you actually asked for:

/g-sesh-goal-add / /g-sesh-goal-rem — record or remove a session-scoped goal (surfaced in gald3r status as it works) without creating a task or touching project state.
/g-codename — add, list, resolve, or remove entries in this project's internal codename/nickname registry.

Cross-project coordination (see Multi-Project Coordination):

/g-wpac-read — triage every incoming item from a linked project's inbox in one pass.
/g-wpac-order — push a task from a parent project down to one or more children.
/g-valk-ask — ask another project a grounded, cited question over the live Valkyrie connector, queuing it if that project is offline.

Full inventory¶

The Commands reference lists every shipped command, grouped by category, with its full description, arguments, and the exact /g-* / @g-* syntax for each supported platform. Use that page as your search target; use this one to get oriented first.