alexs-rig

Plugin alexs-rig in AlexBodner/alexs-rig. Commands read from the repo's manifest — 1 month ago.

Skills (15)

• alex-api

  Design public API surfaces from the caller's side — one obvious entry point, names a user would guess, and the same object serving related workflows. Use when adding or reviewing a class, function or module that library users will call.

• alex-content

  One entry point for outward-facing work — promo video, social post, blog, docs page or figure. Classifies, routes to the skill that owns that format, and applies the gates content actually needs: the claim before pixels, a cheap preview before the expensive render, and approval before anything is pu

• alex-distill

  Shrink L0 when it overflows its token budget — list the largest principles/pending entries and forget or close them at the source. Never silent-truncate. Use when l0-regen reports OVERFLOW or L0 feels bloated.

• alex-docs

  Craft rules for explanatory documents — reports, research write-ups, PR descriptions, findings. Lead with what was done and what came of it, and write one standalone narrative rather than an accretion of replies. Use when writing or restructuring a document someone else will read.

• alex-experiments

  Hygiene for machine-learning training runs and comparisons — what to save while a run is in flight, how to keep runs from overwriting each other, and what makes two arms actually comparable. Use when launching training, setting up an evaluation, or comparing model variants.

• alex-graph

  Incrementally update the codebase graph for only the files changed since the last build (git-tracked), instead of a full rebuild. Refreshing an existing graph runs without asking; only the first build of a repo asks. Use when the graph is stale after edits or a merge.

• alex-loop

  One entry point for a whole task — classify it, research if needed, plan, attack the plan adversarially, then hand off to the right build skill and surface only the polished result. Three gates: the plan, the PR, and any expensive compute. Use to take a task end to end; the individual skills stay ca

• alex-memory

  Surgical updates to Alex's Rig L0 memory (principles, progress, pending). Use when parking todos, updating feature standing, or recording a standing principle after a correction.

• alex-mine-corrections

  Turn captured turns into standing rules. Read the corrections inbox, keep the real corrections, cluster them, and propose a few general reusable rules with evidence; nothing reaches L0 without explicit approval. Use when the Stop hook reports captured turns, or when the user asks to mine, review or

• alex-modularize

  Plan where code should live before writing it — responsibilities, files, directories and the move order that keeps PRs clean, in the layout that fits the kind of project (published library, research repo, applied demo). Use before implementing a feature that adds more than one file, before refactori

• alex-outcomes

  Improve the craft skills from what actually shipped and landed well, not only from corrections. Logs shipped artifacts, attaches how each did, and folds repeated patterns from the winners into alex-viz / alex-docs / alex-api / roboflow-release-promo. Use after publishing something, when attaching a

• alex-roboflow-voice

  The Roboflow house voice for anything written under it or alongside it — READMEs, feasibility reports, release copy, personal repos that carry the same tone. Use when writing or reviewing prose that will be read as Roboflow's, outside the docs pages that trackers-docs already covers.

• alex-structure

  Always-on codebase map. Use for architecture, where a concern lives, blast radius, or module relationships. Prefers understand-anything + codemap-py; grep is fallback.

• alex-verify

  Run the project's checks and record a non-blocking PASS/FAIL status. Informational only — never a gate. Use to back "verifiable" with a real check result surfaced at Stop.

• alex-viz

  Craft rules for visual deliverables — annotated video, overlays, figures, demo renders. Show each fact once, render the intermediate signal a decision is made from, and establish media provenance before anything outward-facing. Use when producing or reviewing a visualization.

Hooks (5)

• SessionStart

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/inject_l0.py"

• UserPromptSubmit

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/prompt_l0_miss.py"

• UserPromptSubmit

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/capture_correction.py"

• PreToolUse

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/secret_hygiene.py"

• Stop

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/stop_review.py"

Similar plugins