hsigstad
UserShared infrastructure for AI-assisted research workflows: Claude Code skills, conventions, methodology docs, and tools.
Categories
Indexed Skills (40)
anecdotes
Collect anecdotal news evidence about a project's topic from Brazilian news outlets (or broader for non-Brazil projects) using the newsbr package. Discovers articles, fetches full text, appends to references/news/stories.csv, and writes a search audit log.
check
Audit a project or pipeline against workspace conventions AND propose fixes for everything found. Three modes: docs (doc-contract + source/build naming), ledgers (artifacts.yaml + validation.yaml + .run.json + cross-refs), cite (citation tokens against the registry + anchors + .bib). All three are fast and deterministic. Default runs all three and proposes changes. Use when the user invokes /check, /check <mode>, /check <slug>, or asks to 'lint the docs', 'audit conventions', 'check citations', 'check the ledgers', 'clean up the docs'.
cite-sync
Regenerate each project's docs/refs/manifest.toml from the workspace citation registry. Walks docs/ and paper/ for [ns:key] tokens with external namespaces, looks each up in research/refs/registry.toml, and writes the public-facing subset (title + description, no internal paths) to the project manifest. Use when the user invokes /cite-sync, /cite-sync <slug>, or asks to 'sync the manifest', 'rebuild the refs manifest', 'regenerate citations manifest'.
distill-style
Read raw entries in research/rules/writing_feedback_log.md, cluster them by tag, propose concrete edits to writing_style.md (general baseline + workspace overlay + per-section files), and let the user approve a batch. Flips status of consumed entries to distilled or rejected. Use when the log has accumulated enough raw entries to be worth turning into rules.
findings-audit
Audit a project's empirical findings against external evidence — anecdotes (news, interviews, court opinions) and aggregate reports (CNJ, IPEA, OECD, FMI). Flags counter-evidence to load-bearing interpretations and surfaces patterns no current finding addresses. Use when the user wants a reality check on findings, before writing up, or after a corpus refresh.
gmail
Draft an email for Henrik in his house style AND record the draft so the sent-vs-draft diff can teach the style. Use when any session drafts/prepares an email for Henrik to send (to a coauthor, colleague, admin, etc.), or when the user says 'draft an email', 'reply to <person>'.
iat
Audit Python scripts for compliance with the Inline Audit Trail (IAT) convention — INTENT, REASONING, ASSUMES, SOURCE comments and validation guards — AND propose the comment text to add. LLM-driven, judgment-heavy, slower than /check. Use when the user invokes /iat, /iat <script>, /iat <slug>, or asks to 'check IAT compliance', 'audit comments in source/', 'add missing IAT comments'.
memory-audit
Audit Claude's persistent auto-memory (MEMORY.md + memory files) against current workspace state; propose stale-entry updates, merges, and deletions for user approval, then apply the approved batch. Use periodically (e.g. when MEMORY.md grows past ~30 entries or quarterly) or when recalled memories keep contradicting the repo.
new-pipeline
Scaffold a new data pipeline under pipelines/ with the pipeline directory structure and required docs. Use when the user wants to create a new data processing pipeline.
new-project
Scaffold a new research project from materials the user has placed in a folder under projects/. Reads existing files to extract context instead of asking questions. Use when the user wants to create a new research project.
otter
Pull recent Otter.ai meeting-summary emails from Gmail, route each to the right project, and save under docs/meetings/. Use when the user runs /otter or asks to 'grab the otter notes' / 'sync otter meeting summaries' / 'update meeting notes from otter'.
site
Generate the static HTML site for a research project. Creates source/site/build_all.py and templates if they don't exist, then builds the site. Use when the user wants to create or rebuild a project site.
tactiq
Pull recent Tactiq meeting transcripts from Google Drive, route each to the right project, and save as raw markdown under docs/meetings/. Use when the user runs /tactiq or asks to 'grab the tactiq notes' / 'sync meeting transcripts' / 'save the recent tactiq transcripts'.
validate-section
Run AI validation on one subsection of the paper: identify its backing scripts, run the applicable AI checks, and record results in the validation ledger. Use when the user asks to validate a section or asks what AI has verified for a section.
backup
Back up expensive-to-regenerate pipeline/project artifacts to Dropbox with the right format and location. Use when the user wants to offload, back up, or archive data to Dropbox — a corpus, a cache, scraped raw, or a whole pipeline's build outputs.
data
Search the data catalog, variable dictionary, and data linkages for datasets, variables, or data sources. Use when the user asks about available data, what variables exist, or how datasets connect across projects.
diarios
Reference for the shared diarios Python module — API lookup, usage patterns, and gotchas. Use when writing code that involves court data, legal text parsing, Brazilian administrative data cleaning, or when checking if a utility already exists before writing a new one.
dropbox
Browse, upload, or download files on Dropbox via rclone. Use when the user wants to share files with collaborators or check what's on Dropbox.
fetch-annotations
Pull hypothes.is annotations left by coauthors on a project's rendered site, map each one back to its source .md file, walk the user through applying (or skipping) each as an edit, write a response log, and optionally post condensed replies back as threaded annotations on hypothes.is. Use when the user wants to incorporate feedback left as web annotations.
findings
Populate, extend, refresh, or audit a project's docs/findings.md — a curated index of headline empirical findings and interpretations with confidence tags and standardized source footers. Use when the user wants to draft a new findings doc, append new entries from recent build artifacts or audit JSONs, refresh load-bearing numbers against the current data, or check completeness.
handoff
End-of-session: commit and push all changes across projects touched in this session. Write a handoff note only when there's a cognitive thread the rest of the system doesn't capture (mid-stream WIP, noticed-but-not-acted-on, partial verification). Use when the user is done working and wants to wrap up.
hypothesis
Populate or expand a project's docs/hypotheses.md (or docs/hypotheses/ folder) — testable predictions linking theory (docs/theory.md or docs/literature.md) to evidence and the project's empirical design. Use when the user wants to draft, update, or audit hypotheses.
idea
Create a new research idea with YAML frontmatter and add it to ideas/index.md. Use when the user wants to record a new research idea.
inbox
Send a message from one Claude session to another concurrent session — typically a sandboxed session asking the host (non-sandboxed) session to do something the sandbox can't (write via a read-only connector, run a host-only tool), or leaving a note for a named session. Delivery is automatic via the inbox hook. Also covers the rare container (e.g. ~/me) that can't reach the shared inbox and needs a manual relay.
institutions
Populate or audit a project's docs/institutions.md using the shared institutional reference (research/institutions/brazil/ for Brazil projects). Bootstraps from shared topic files and artigos.db; marks gaps for manual curation.
literature
Discover academic literature for a project: search APIs, curate relevant papers, write literature.md + bib entries, download PDFs. Use when user wants to find, update, or expand a project's literature review.
llmkit
Reference for the llmkit LLM extraction framework — cache design, Pydantic validation, audit workflow, and per-project setup. Use when writing code that does LLM-based structured extraction, setting up new extraction tasks, or working with cached LLM outputs.
meet
Create a Google Meet meeting as a Google Calendar event. Use when the user wants to schedule a meeting with a Google Meet link, e.g. 'set up meet with Darcio Tue 2pm' or '/meet Gui Friday 10am'.
next
Run one iteration of the project analysis loop: accept (or propose) the next analysis, write the script in the right location with IAT, run it, then propagate to the right docs. Use when the researcher says 'next, do X' or asks 'what's next?'.
questions
Populate or update a project's docs/questions.md — the 3-5 policy-relevant research questions the project answers, each synthesizing evidence from multiple hypotheses. Use when the user wants to draft, update, or audit questions.md.
style-check
Fast deterministic prose-style checker. Wraps research-kit/tools/style_lint.py for file-scoped use. Use when the user invokes /style-check <file>, asks to "check style", "scan for AI tells", "polish before sending", or wants a quick mechanical pass before submission. For LLM whole-document review use /style-review; for section-scoped pre-submission validation use /validate-section.
style-review
LLM-based prose review against the empirical-economics writing style guides. Runs the mechanical linter first, then reads the file against the general baseline and any relevant per-type guides. Use when the user invokes /style-review <file>, asks for a "thorough style review", "polish this for submission", "review for clarity", or wants prose feedback on a paper draft / important email / chapter. For fast deterministic AI-tell scanning, use /style-check; for section-scoped pre-submission validation with ledger writes, use /validate-section.
style-revise
Revise a rough draft paragraph by fixing idea flow before wording, then offer alternative shapes. Use when the user shares unpolished prose and asks for comments, says a paragraph "feels awkward", wants "better flow", restructuring, wording alternatives, or help finding the right shape for ideas they've already drafted. For auditing a near-final document against the style guide use /style-review; for a fast mechanical AI-tell scan use /style-check.
theory
Populate or expand a project's docs/theory.md — a structured inventory of formal theoretical frameworks relevant to the project, with testable predictions tied to the project's identification strategy. Use when the user wants to draft, update, or audit theory.md.
Send or read WhatsApp messages. Use when the user wants to message a contact or group, e.g. '/whatsapp Andrei hey, check the new data'.
zoom
Create a Zoom meeting as a Google Calendar event. Use when the user wants to schedule a meeting, e.g. 'set up zoom meeting with Darcio Tue 2pm'.
referee-panel
Run a finding-level mechanical auditor panel over a project's paper build: ~12 single-failure-mode lenses (identification, claim-evidence, numerical, model-equation, theory-logic, robustness, sample-construction, abstract-conclusion, literature, external-validity, measurement-validity, power/multiple-testing), each emitting structured findings under a cannot_verify-or-cite contract, then an adversarial verify pass, then an editor memo. Use for a pre-submission red-team of a paper, complementary to /findings-audit (evidence) and the community-persona referee reports.
howto
Print Henrik's one-page Claude Code cheat sheet (situation -> command).
drive
Browse, upload, or download files on Google Drive via rclone. Use when the user wants to access or share files on Google Drive.
outlook
Convert Outlook .eml email exports to clean markdown for project docs. Use when the user wants to save email exchanges to a project's docs/emails/ directory.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.