sirmaelstrom
@sirmaelstrom
Public Skills
slim-review
by sirmaelstrom
"Run the slim PR-review loop — two external reviewers (GPT-6 Sol and Astra, plan-covered) at a PR boundary, findings posted as line-anchored GitHub review comments, then confirm or refute each and reply. Trigger on '/slim-review', 'slim review', 'review this PR before merge', or at any PR boundary in a burn-down. The LIGHT tier: two lenses, no synthesis. NOT for converging a reviewed PR (babysit), in-session multi-reviewer passes ('/review'), or complex PRs (review-council)."
babysit
by sirmaelstrom
"Converge a PR's paired review under bounds: claim the head, judge findings, fix, push, until a clean pass with threads resolved and CI green, or a blocked exit naming what is owed. Trigger on '/babysit', 'babysit this PR'. Managed repos only; never merges."
burn-down
by sirmaelstrom
"Run an ordered queue of roadmap quests to landed in one or more sittings, holding the disciplines that keep the board honest — pickup flip, receipt per stop, ripple closeout, side-thread harvest, owed-step receipts. Trigger on '/burn-down', 'burn down the queue', 'work the queue', 'run the burn-down', 'clear the frontier', 'work these quests to landed'. For resuming a SINGLE quest use pickup instead; this is the multi-item runner that calls it."
codex-delegate
by sirmaelstrom
"Use when offloading a bounded audit, broad search, extraction, or mechanical edit to a plan-covered CLI agent. Trigger on 'delegate' or 'offload'. Not for PR review (use slim-review), work needing this session's decisions, or a repo you are mid-editing."
spec
by sirmaelstrom
"Generate a specification. Trigger: '/spec', 'spec this', 'write a spec for', 'auto-spec'. READ THE FULL SKILL before executing — depth selection, pipeline stages, and review gates are defined below."
pickup
by sirmaelstrom
"Resume the current project's top roadmap quest from the Spine — read its resume note + cartridge artifacts, claim the quest on the board, and continue the work. Trigger on '/pickup', 'pick up where I left off', 'continue the quest', 'resume the roadmap work' (the work-resume sense; Claude Code's built-in /resume resumes SESSIONS — this resumes WORK). Optional arg: a quest short-id or project name to pick up something other than the default."
session-chain
by sirmaelstrom
"Use when a long-running Claude session needs a durable successor in herdr. Trigger on '/session-chain', 'rotate this session', 'chain to a fresh session', or 'spawn a Claude session in herdr'. NOT for in-turn delegation (Agent tool) or codex lanes (codex-delegate / lane.mjs)."
failure-audit
by sirmaelstrom
"Use when auditing newly archived agent-session transcripts for delta failure modes. Trigger on '/failure-audit', 'failure audit', 'audit the new transcripts', 'delta audit'. NOT for scheduled runs or continuous scanning; use operator-pulled spot checks."
commit-msg
by sirmaelstrom
"Use when composing a git commit message with shell-active characters — backticks, markdown links, embedded quotes, multi-line code blocks — anything a HEREDOC could re-parse. Invoked by '/commit-msg', or whenever the drafted message contains any of these. Writes the message to a file (.git/COMMIT_EDITMSG, or a scratchpad path in a linked worktree) and commits via 'git commit -F' so the shell never sees the body."
handoff
by sirmaelstrom
"Compact the current conversation into a handoff document so a fresh agent can pick up the work without losing intent. Trigger: 'hand off', 'handoff to a fresh agent', 'pass this to a new window', 'compact context', or when the context window is heavy mid-task and you need to continue in a fresh session. Also use the DIY sub-agent pattern: handoff out, work in another window, hand back."
wizard
by sirmaelstrom
"Generate an interactive bash wizard that walks a human through steps only they can perform. Trigger: 'wizard for X', provisioning infrastructure, credentials/CI-secrets setup, unfamiliar third-party dashboards, one-off migrations or cutovers. Not for steps the agent can do itself."
blast-radius
by sirmaelstrom
"Find what a change could break somewhere else before it ships, beyond the diff, and prove the one fact it's safe because of by running real code instead of writing it up. Trigger: 'blast radius of X', 'what could this break', 'impact analysis', reviewing a small diff you don't trust, or pre-PR on a change that touches a contract, wire format, or shared surface."
unslop
by sirmaelstrom
"Cut AI tells from prose — puffery, filler, hedging, chatbot phrases, weak verbs, metaphor jargon — and put a human voice back in. Trigger: 'unslop', 'deslop', 'make this read human', or before publishing docs, README prose, PR bodies, briefs. Not for code or quoted material; domain terms defined in UBIQUITOUS_LANGUAGE.md are exempt from the jargon cut."
parallel-explore
by sirmaelstrom
"Fan out a design or problem space into multiple intentionally different agent runs, then synthesize into a ranked decision. Use when the user says 'explore alternatives', 'parallel explore', 'design it twice', 'what are my options', wants to compare approaches, or needs to make an architectural decision between competing designs. Do not use for reviewing an existing implementation, or when the open question can only be settled by running code."
review
by sirmaelstrom
"Review a target with an adaptive multi-reviewer pipeline: GitHub PR, local branch (pre-PR), working tree, file/directory, or plan/spec. Triggered by '/review', '/review <PR-URL>', '/review #<num>', '/review branch[:<name>]', '/review wt', '/review file <path>', '/review plan <path>', or 'review my branch / plan / changes'. Not for a single-lens pass at a PR boundary — use slim-review for that."
spec-validate
by sirmaelstrom
"Validate a workshop spec's quality. Trigger: 'spec-validate', 'validate spec', 'check workshop', 'is this ready?'. READ THE FULL SKILL — validation checks and script usage are defined below."
wait-what
by sirmaelstrom
"Restate the last explanation in short declarative sentences grounded in the project's own vocabulary. Trigger: 'wait, what?', 'I don't follow', 'what do you mean', 'explain that again', 'come again', or any signal the operator lost the thread."
grill-me
by sirmaelstrom
"Stress-test a plan or idea. Trigger: 'grill me', 'poke holes', 'what am I missing', or wanting to pressure-test before committing to a spec."
diagnose
by sirmaelstrom
"Diagnose a failing behavior by forcing environmental, process, config, and code-path hypotheses into an explicit verification flow before making changes. Use when something is broken, crashing, not working as expected, behaving strangely, or when the user says 'diagnose', 'debug this', 'why is this broken', 'figure out what's wrong'. Also trigger proactively when investigation starts drifting toward code edits without evidence."
improve-architecture
by sirmaelstrom
"Use when the user says 'review architecture', 'find code smells', 'what should I refactor', 'improve structure', or wants to make a codebase more agent-friendly and testable."
audit-skills
by sirmaelstrom
"Score skills against the evaluation rubric and write results to skills.db. Use when the user says 'audit skills', 'score my skills', 'skill health check', 'rate skills', 'how are my skills', or wants to assess skill quality across plugins. Also trigger when a significant skill change just shipped and needs reassessment. NOT for running skills against test cases to improve them (that's eval-loop) or grading spec quality (that's spec-validate)."
execute-wp
by sirmaelstrom
"Execute a work package from a spec. Trigger: 'execute WP-01', 'run wp', 'execute wp', or referencing a specific work package. READ THE FULL SKILL — the execution lifecycle (read orchestrator → execute → append progress log) is defined below."
ubiquitous-language
by sirmaelstrom
"Extract domain terminology into a glossary. Trigger: 'domain language', 'glossary', 'terminology', or when ambiguous terms cause confusion in a project."
chart
by sirmaelstrom
"Map a long-running effort as a decision tree on disk — a destination, the open questions beneath it, and one question resolved per session. Trigger on '/chart', 'chart this effort', 'map out this decision', 'what should I decide next', 'advance the chart', 'record an answer'. This is DECISION MAPPING, not data visualization — for a bar chart, graph, plot, heatmap, or dashboard use the dataviz skill instead."