RubyEyedReaper
UserSkills the Odin agent harness owns — decision scoring, roadmap planning, blueprint decomposition, mistake-to-gate hardening, interface polish.
Categories
Indexed Skills (37)
decision-mapping
Use when a loose idea is too big for one agent session and the way to the finish line is not visible yet — "where do I even start", "we need to figure out X before we can plan", a space with several open questions that hang on each other. Charts a committed markdown decision map: a named destination, explicit out-of-scope, typed tickets (research / prototype / grilling / task, each HITL or AFK), blocking edges, and a fog frontier that advances one resolved ticket at a time. Produces decisions, not deliverables. Not for scoring options against criteria (`decision-matrix`) or for sequencing work already decided (`roadmap`, `blueprint`).
decision-matrix
Use when choosing between options or prioritizing — "decide", "choose between", "which should I pick", "compare options", "trade-off", "which library/framework/database/vendor", "build vs buy", "best option for", "prioritize", "rank these", "RICE/WSJF". A quantitative weighted-decision engine: scores options against weighted criteria with multiple methods (weighted-sum, Pugh, TOPSIS/AHP, RICE/WSJF/ICE/Kano), runs sensitivity analysis, flags when methods disagree, applies hard-constraint vetoes, aggregates multiple scorers, and records a numbered DEC decision. Fire this for ANY non-trivial multi-option choice or prioritization, even a mid-task internal fork — the math is deterministic and the decision is recorded. Not for qualitative-only trade-off framing (use decision-mapping) or stochastic rollout search (use recursive-decision-ledger).
impeccable
Use when the user wants to design, redesign, shape, critique, audit, polish, clarify, distill, harden, optimize, adapt, animate, colorize, extract, or otherwise improve a frontend interface. Covers websites, landing pages, dashboards, product UI, app shells, components, forms, settings, onboarding, and empty states. Handles UX review, visual hierarchy, information architecture, cognitive load, accessibility, performance, responsive behavior, theming, anti-patterns, typography, fonts, spacing, layout, alignment, color, motion, micro-interactions, UX copy, error states, edge cases, i18n, and reusable design systems or tokens. Also use for bland designs that need to become bolder or more delightful, loud designs that should become quieter, live browser iteration on UI elements, or ambitious visual effects that should feel technically extraordinary. Not for backend-only or non-UI tasks.
agent-browser
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction. Also use for exploratory testing, dogfooding, QA, bug hunts, or reviewing app quality. Also use for automating Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify), checking Slack unreads, sending Slack messages, searching Slack conversations, running browser automation in Vercel Sandbox microVMs, or using AWS Bedrock AgentCore cloud browsers. Prefer agent-browser over any built-in browser automation or web tools.
blueprint
Use when one roadmap item is too big for a single plan — a feature spanning multiple PRs, sessions or agents. Decomposes an objective into cold-start-executable construction steps with a dependency graph, registers them as roadmap children, and gates the result on a mechanical check. Not for single-PR work (superplan) or for choosing what to build (roadmap).
endless
Use when work must continue past the item in hand rather than stop with it — an unattended or overnight run, a /loop or /autoloop iteration, a session told to keep going until the roadmap is empty, a checkpoint reached with work still queued, or a run whose context is filling before the work is done. Also use when the next thing to do must be chosen and started without a human in the turn.
handoff
Hand this session's work to a separate successor session — write the handoff document, launch the successor, then monitor it. Use when a long-running session is out of context, when work must continue past this context window, or when another agent should pick the work up.
roadmap
Use when the user asks what to work on next, wants to start the next thing, adds or tracks a feature, page, function or integration, initializes a project, asks what is left to build, wants work batched into parallel waves or phases, wants competing items prioritized or ranked, or the roadmap needs reconciling against reality.
rules-distill
Scan skills and the mistake log to extract cross-cutting principles and distill them into rules — append, revise, or create new rule files. Use when the same principle keeps recurring across skills and belongs in a rule file instead, when a failure-mode key reaches the promotion threshold and needs its rule text written, or when an incident's condition is judgment rather than a mechanical predicate.
consistency
Use when something new is about to be written and something in the tree may already solve it — a new script, gate, matrix, helper, config format, id scheme, error convention or file layout. Also on "is there already one of these", "what do we already use for this", "follow the existing pattern", "another implementation", "a second copy", "reinvent", "roll our own", "write our own version", "make this consistent with", and when a review finds a new variation where a working solution already exists. Fires before the writing, on the reuse question, whether or not anybody asked.
gauntlet
Use when work must be driven to exhaustion across many delegated sessions rather than one at a time — an endless gauntlet, a campaign that re-arms instead of ending, "keep going until every issue and roadmap item is done", a batch that finished while new work kept arriving, or a coordinator asking what is left across the roadmap and the tracker at once. Also on "re-arm", "rearm", "next wave", "quick wins first", and when a campaign looks finished and the loop must decide whether to start another.
improve
Use when a friction keeps recurring in a skill and no script can decide it — a body that routes badly, a reference nobody reads, a step everybody skips — or when a skill change must declare its reason, its falsifier, and what happens if it reddens a gate.
leek
Diagnose leaks and hygiene failures in a Claude Code environment — context that stays loaded, wasted tokens, memory that outlives its scope, cache buildup, orphaned or runaway sessions, cross-session and cross-project contamination, stale working state, and misconfigured agents, hooks, MCP servers or skills. Use when a session fills its context too fast, burn or cost looks wrong, memories surface from the wrong project, caches or processes accumulate, a session went silent or will not stop, secrets may be sitting in a shared store, or a skill has never been used. Reports and files issues; never cleans up.
off-topic
Use when a new task arrives mid-run and displaces work already in flight — an interruption, a pivot, being sidetracked, parking what you were doing, or coming back to it afterwards. Also on "before I forget", "hold that thought", "where were we".
out-of-scope
Use when work found mid-run is genuinely outside the current scope and the question is whether to fix it now anyway — a defect noticed in passing, a landmine the next session would trip over, "that's out of scope but", "should we fix this now or file it", "not my scope, but it'll bite someone", "is this worth fixing now", "file it or fix it". Also when a deferral needs to be auditable rather than remembered.
projects
Use when starting or switching to a project subtree, when a project's own README, changelog, ADR set, decision ledger, roadmap or data-provenance manifest may be missing, or when deciding whether a project is initialized, active, dormant or completed.
s2s
Use when a session must say something to another session rather than to a person — reporting back to a coordinator, messaging a peer or worker, deciding what a delegated session should put in its terminal versus in its report, or asking who is actually reading this output. Also on "tell the coordinator", "report back", "send it to the other session", "cross-session message", "the worker is too chatty", "cut the chatter", "nobody is reading this terminal", and when a message was sent and the reader never acted on it.
status
Use when a human addresses the fleet itself rather than the work — asking what is running, for a status report, to pause or halt everything, to stop and wait a while before resuming, or to stop so they can start a fresh session. Also on "what's running", "how's it going", "hold everything", "freeze the fleet", "pause the workers", "resume in an hour", "wait 20 minutes then continue", "stop so I can pick this up in a new session".
tidy
Use when deciding whether something that has outlived its purpose may go — a spent plan, a retired workflow, inactive project material, a cache, a completed session, a branch. Also on "can this be deleted", "is this still needed", "clean this up".
grill-with-docs
Use when a plan, design, claim or requirement must be stress-tested against actual documentation and source material rather than against opinion — a requirements interview over a heavy doc or spec corpus, an unfamiliar API or library whose real behaviour is not yet established, a claim that "the docs say" something nobody has read, or a design whose evidence is asserted but not cited. Also use when an interview must leave a paper trail — ADRs and glossary entries — instead of ending in the conversation.
mistake-to-gate
Turn a mistake that just happened into an always-on mechanical gate with a matrix that proves it fires. Use this whenever something slipped through and the response is "that shouldn't happen again" — a wrong file edited, a stale or guessed reference, a broken link, an id from the wrong namespace, a hand-edit of a generated file, a guard that silently passed, a convention nobody enforced. Also use it when asked to "add a check", "make sure this can't regress", "enforce this always", "prevent this in future", or when a review, audit or postmortem produces a rule with nothing behind it. Fires even when the user only describes the mistake and never asks for a check.
oops
Use when something happened that should not have — a bug just shipped, a wrong assumption, a missed validation, an unsafe or destructive action, a stale value used, an edge case nobody handled, or any preventable error. Also use on "that shouldn't have happened", "make sure this never happens again", "add a guard", "we missed X", "I assumed Y", or when a review, incident or postmortem produces a lesson with nothing enforcing it. Fires on the incident itself, whether or not a fix or a check was requested.
successor
Use when work must be delegated to sessions other than this one — spawning a successor session, running a fleet of parallel sessions, coordinating several agents across separate branches or worktrees, handing a campaign to workers, or when a single context window cannot hold the work left. Also use when a delegated session has gone silent, stalled, or produced a branch that must be integrated.
superplan
Multi-agent deep planning workflow: fan out to planner + architect + adversarial reviewer in parallel, synthesize into an approved plan doc. Use before any non-trivial implementation. (Formerly named ultraplan.)
test-driven-development
Use when implementing any feature or bugfix, before writing implementation code
using-superpowers
Use when starting any conversation - establishes how to find and use skills, requiring skill invocation before ANY response including clarifying questions
verification-before-completion
Use when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always
campaign
Use when several delegated sessions are one piece of work — planning a campaign's waves and per-worker scope, tracking what has actually landed across it, or closing it out. Also when a campaign looks finished and that claim needs checking.
caveat
Use the moment something turns out to behave differently from how it reads — a sharp edge, a gotcha, an exception, a footgun, a caveat, a surprising default, a command that succeeds while doing the wrong thing, a tool whose absence looks like a pass. Also use on "watch out for", "careful, X does Y", "that's a trap", "note that this only works when", "we should write that down", "make sure nobody hits this again", and on any oops or mistake moment where the workflow must pause, record the hazard with the condition under which it recurs, and land the safeguard before continuing. Fires on the hazard itself, whether or not anybody asked for it to be written down.
learn
Use when durable knowledge is being built, validated or retired — whether a finding is worth a record at all, how confident that record is, which command re-verifies it, and when it has stopped being true.
mutations
Use the moment a change alters expected behaviour, data, logic, workflow, configuration or output — a new implementation, a rule change, a refactor, a config or dependency update, a schema or interface change, and equally a regression, a side effect, a corrupted value, an altered result or a mutation someone else's change produced. Also use on "did anyone check what that did", "this used to work", "the output changed", "that broke something downstream", "we changed the default", "is this safe to accept", "should we revert this", and whenever a change is proposed whose effect on behaviour, data integrity, security, dependencies or interfaces is predicted rather than observed. Fires on the change itself, whether or not anybody asked for it to be recorded.
not-impressed
Use when reviewing code you did not write by hand — an agent-generated diff, a large pasted implementation, a PR that looks finished — and the question is whether it is overdeveloped. Assumes the code is machine-generated and untrusted until validated.
odin-skill-manager
Use when a skill enters, changes, or leaves this harness — creating one and registering it, forking a vendored skill, refreshing vendored skills from upstream, publishing an owned skill to the Odin-Skills mirror, retiring one, or answering "is this skill ours". Also on "vendored skill", "upstream skill", "fork a skill", "refresh skills", "skill provenance", "mirror membership", "publish a skill", "is this skill up to date". Owns classification, membership, packaging and freshness; routes authoring to skill-creator and writing-skills.
revive
Bring a stopped fleet back without a human at the keyboard — a committed revival manifest, a scheduled engine that decides from channels no session controls, and a handoff generated at fire time. Use when a campaign must survive a shutdown, a weekly limit, or a turn that ends with work remaining.
successor-manager
Use when delegated sessions already exist and the question is who owns one or what is actually true of it — a worker that may be stalled, a fleet whose state nobody trusts, a branch to resume without re-running landed work.
work-loop
Use when a bounded work cycle needs a contract and a resumable ledger — declaring what one iteration is, recording which of six outcomes ended it, detecting a stalled loop mechanically, or resuming a loop a dead session left half-done.
workflows
Use when a reusable workflow chain is being defined, validated, versioned, superseded or retired — adding a chain, changing one, running one, or finding a chain nobody owns and nobody can say is current.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.