dgilford
UserSkills & reviewer agents for AI-first climate science — built and used by a PhD atmospheric scientist
Categories
Indexed Skills (18)
ai-review
SLASH COMMAND — type /ai-review for a comprehensive senior-engineer review of a project or repository. Orchestrates a parallel fan-out across correctness, staleness, over-baking, and claim rigor by DELEGATING to the tools that own each lane, then adds the three lanes nothing else covers — gap/opportunity hunting, grounded novel ideation, and a single prioritized synthesis. Report-only by default; `--fix` opts into HIGH-confidence deterministic repairs.
create-alert
SLASH COMMAND — type /create-alert to author a scheduled Slack alert. Grills a fuzzy "tell me when X happens" ask into a testable trigger spec, dry-runs it live (no Slack send), and — only after you explicitly sign off on the synthesized spec — creates a claude.ai cloud routine that evaluates the trigger on a schedule and messages Slack when it fires. Create-only; never creates before sign-off. `--spec-only` stops at the spec without creating anything.
figure-review
Audit a scientific figure for publication-readiness: colormaps, uncertainty, axis labels, caption completeness, and claim support. Panel-aware — judges one panel of a composite against what its shared caption and sibling panels supply, and raises a likely-deliberate choice as `by-design?` instead of a defect. Use this whenever the user shares or references a figure, plot, panel, or colorbar for a paper, report, or brief and wants it checked, reviewed, or made publication-ready — even if they just say "does this figure work?" or "review my plot." Emits a per-criterion report and archives it to .ai/reviews/; never silently rewrites plotting code.
handoff
Create or update a durable project handoff for the next AI agent/session. Use when ending a coding/research session, switching agents, compressing context, or preparing another Claude/ChatGPT/Codex session to continue the work.
overbaked
Audit a document, plan, or code for over-engineering, verbosity, and scope creep. Use when you've just written something and want a point-by-point check before finalizing.
repo-init
SLASH COMMAND — type /repo-init to scaffold a new repository (or retrofit an existing one) with a standard project structure. Starts with a short intake grill (--no-grill skips). Research mode by default; --package for a distributable src-layout library. Never overwrites existing files; --dry-run previews.
unstale
Detect and repair staleness residue in Python library code and notebooks — dead imports, dead code, resolved TODOs, stale comments/docstrings, and HANDOFF blockers. Use this whenever you want to clean up after a refactor, fix a comment that no longer matches the code, remove unused imports, check for leftover TODOs, detect doc drift, or tidy a notebook before sharing or submission — even without the word "unstale." Default mode auto-detects library .py scope and runs ruff + vulture (deterministic), then LLM semantic checks; never touches notebooks. `--artifact <nb.ipynb>` cleans a notebook as a deliverable (full treatment including dead-code detection). `--exploratory <nb.ipynb>` tidies a working notebook non-destructively (structural + semantic only, no removals ever). Always emits a structured report before any edit.
worklog
Capture a work-log entry to the Notion Work Journal, a remote server cache, and a local mirror. Use whenever the user asks to log, note, record, or journal progress, a decision, or what they just did — and as the capture step invoked by /handoff.
pickup
Pick up work from repo-local handoff state. Use when starting a new Claude Code session, switching agents, returning to a project, or asking an AI agent to reconstruct context from prior work. Named `pickup` rather than `resume` to avoid shadowing Claude Code's built-in `/resume` conversation picker.
slack-message
Draft an internal Slack message grounded in current project context and recent workflow. Use when sharing a status update, result, finding, or request with a teammate or internal stakeholder at your org.
tab-setup
Set a unique color and name for this Claude Code tab based on the current working directory. Use "all" to recolor and relabel every active session at once.
commit-batching
Batch a dirty working tree into logical, single-concern commits instead of one monolithic `git add -A`. Use whenever you are about to commit and more than one file or concern has changed — survey every change, group by concern, stage each group by explicit path, write a focused message per commit, then push only if authorized.
evolve-claude-md
Update CLAUDE.md — or the canonical AGENTS.md it redirects to — with durable knowledge from the current session. Use at the end of a working session to promote new findings, renamed variables, refined definitions, or corrected assumptions into the permanent project record.
lit-review
Search and synthesize scientific literature from Zotero, arxiv, bioRxiv, Google Scholar, and Consensus. Use when framing a research question, designing methods, interpreting results, mapping the field landscape, or identifying future directions.
pathfinder
SLASH COMMAND — type /pathfinder to get a navigable map of every skill and subagent and when to reach for each. Resolves the reviewer-2-vs-panel decision. Use when unsure which skill or agent to invoke, or when orienting a new session.
reviewer-2
Adopt a critical-reviewer stance to stress-test a claim, result, or manuscript section. For each claim: identify the baseline being compared against, name the counterfactual, surface plausible alternative explanations, and check that stated uncertainty is consistent with the strength of the claim. Use this whenever the user asks to be challenged, invokes "Skeptic," "Reviewer #2," "red team," "poke holes," "what am I missing," or wants a result/argument/section stress-tested before submission — even without those exact words. Reports concerns; does not rewrite the user's argument.
write-new-skill
Create new Claude Code skills with proper structure and progressive disclosure. Use when user wants to create, write, or build a new skill.
grilling
Grill the user relentlessly about a plan or design, one decision at a time, until you reach shared understanding. Use when the user wants to stress-test a plan before building or uses a "grill" trigger phrase — and proactively, before any step that changes running state, allocates a shared resource (GPUs, ports, memory), edits an interdependent config, or could invalidate an earlier decision. Adapted from Matt Pocock.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.