harness-configs

Use when the user wants to apply leave / absence in Workday (sick leave, casual leave, planned leave, comp off, etc.) for one or more days, invoked as /apply-workday-leave. Review-first — it shows the absence plan (dates, type, hours) and waits for confirmation before submitting, then reconciles the timesheet so the leave day carries no project hours. Drives the user's real Chrome via the chrome-cdp CLI (logs in through login-microsoft-sso).

Use when the user wants to review and approve their pending Workday "My Tasks" approvals via the browser — invoked as /approve-workday-tasks. Lists pending items and approves only the ones the user explicitly selects. Drives the user's real Chrome via the chrome-cdp CLI (logs in through login-microsoft-sso).

Use when the user wants to fill in their Workday timesheet for the current week (hours per weekday against a project), invoked as /fill-workday-timesheet. Review-first — shows the week's planned hours and waits for confirmation before saving. Drives the user's real Chrome via the chrome-cdp CLI (logs in through login-microsoft-sso).

Use when the user wants a list of their meetings for a week from the Outlook (Microsoft 365) calendar — invoked as /list-week-meetings. Read-only: it reads the week's events and presents them grouped by day (time, title, organizer, online/in-person, status). Drives the user's real Chrome via the chrome-cdp CLI (logs in through login-microsoft-sso, app outlook).

Use to ensure an authenticated browser tab for an app behind your organization's Microsoft (Entra) SSO — e.g. Workday, Engage, Outlook — driven by the `chrome-cdp` CLI. It drives the user's real, already-logged-in Chrome, so it types no credentials; it navigates to the app and, if redirected to a sign-in page, clicks that app's SSO entry, which bounces back in via the live session. A building block for the automation skills — invoke it first (with the target app) so they operate on a logged-in tab. Invoked as /login-microsoft-sso.

Use when the user wants to record an activity in Engage (the org's activity/points platform) — e.g. a billed work week, a phone interview, or an attended course/class (e.g. ImprovingU) — invoked as /record-engage-activity. Fills the Add Activity form (category, type, date, quantity, notes) and submits only after user confirmation; can derive a class's date/hours from Outlook. Drives the user's real Chrome via the chrome-cdp CLI (logs in through login-microsoft-sso).

Assist any Indian individual with preparing and e-filing an Income Tax Return (ITR-1/2/3/4) on the e-filing portal (eportal.incometax.gov.in), under EITHER the old or new tax regime. Use WHENEVER the user mentions filing taxes in India, ITR, income tax return, 26AS, AIS, Form 16, old vs new regime, 115BAC, Form 10-IEA, 80C/80D/HRA/home-loan/NPS/80G deductions, 44ADA/44AD presumptive, self-assessment/advance tax/234B/234C, TDS reconciliation, capital gains on Indian shares/mutual funds/property, or crypto/VDA (115BBH/194S) — even if they don't name the form or regime. Covers gathering and reconciling income documents, comparing both regimes to pick the cheaper one, choosing the form, computing tax, filling the portal schedule-by-schedule, fixing validation defects, and guiding payment and e-verification. India personal income tax only — not US/UK/other-country tax, GST, TDS-return (24Q/26Q), or company returns.

Apply Edward Tufte's information-design principles to ANY information display — slides/decks, documents, blog posts, dashboards, HTML artifacts, tables, diagrams, emails — when creating one with clarity as a goal or reviewing one. Use on triggers like "apply Tufte", "make this clearer", "less cluttered", "too busy", "chartjunk", "improve information density", "review this slide/report/dashboard/diagram/table for clarity", "is this chart misleading". A principles + review lens, not a builder — chart construction mechanics (marks, palettes, axes) belong to a dedicated dataviz skill when one is available; deck/doc file mechanics to pptx/docx; visual styling direction to a frontend-design skill when one is available.

Mine the knowledge-base wiki for strong, defensible content ideas — blog posts, conference talks, internal sessions, threads. Use when the user says "ideate", "what should I write about", "find content ideas in my wiki", "blog/talk ideas", or wants to turn collected knowledge into publishable content. Produces a scored shortlist with outlines in output/ and maintains an ideas backlog in wiki/synthesis/.

Process raw source documents into wiki pages. Use when the user adds files to raw/ and wants them ingested, says "process this source", "ingest this article", "batch ingest", "deepen", "I added something to raw/", "I clipped some articles", or wants to incorporate new material into their knowledge base. Sweeps the vault's Clippings/ folder (Obsidian Web Clipper) into raw/ first, enriching metadata on the way.

Sync Readwise highlights and Reader documents into the second-brain vault's raw/ folder in Obsidian-Web-Clipper format. Use when the user says "sync readwise", "pull my readwise highlights", "update raw/ from reader", or wants their reading library reflected in the wiki (triggers: readwise sync, reader sync, import highlights). Sync only — follow with /second-brain-ingest.

Health-check the wiki for contradictions, orphan pages, stale claims, and missing cross-references. Use when the user says "audit", "health check", "lint", "find problems", or wants to improve wiki quality.

Answer questions against the knowledge base wiki. Use when the user asks a question about their collected knowledge, wants to explore connections between topics, says "what do I know about X", or wants to search their wiki.

Resurface knowledge from the second-brain wiki — a daily/periodic review of highlights, concepts, and stale pages, replacing Readwise's daily review. Use when the user says "daily review", "review my highlights", "resurface something", "what should I revisit", or wants spaced-repetition-style engagement with their wiki.

Set up a new Obsidian knowledge base with the LLM Wiki pattern. Use when the user wants to create a second brain, initialize a vault, set up a personal knowledge base, or says "onboard". Guides through an interactive wizard to configure vault name, location, domain, agent support, and tooling.

Use to add LLM-friendly outputs to a Hugo site — /llms.txt and /llms-full.txt indexes plus a per-page markdown twin at <url>/index.md — generated from content so they stay in sync. Triggers "add llms.txt", "make the site agent-friendly", "markdown twin".

Pull the heap/goroutine pprof profiles a CI job captured, separate a real leak from baseline cost, and quantify a fix's before/after delta. Use for "where is the memory/compute going", "is this a leak", "verify the perf fix", "compare profiles from the latest CI run", or when a Go service's CI run uploads pprof artifacts. Generic to any Go project whose CI captures /debug/pprof dumps.

Run a Prometheus TSDB snapshot that a CI job uploaded inside a local Prometheus container and query it — for before/after performance comparisons across legs and leak-vs-constant-offset questions a point-in-time pprof can't answer. Use when a CI run uploads a prometheus data backup and you need time-series (latency, goroutines, RSS, CPU) over the whole run, or to compare metrics before-vs-after across CI legs. Generic to any project whose CI backs up its Prometheus data dir.

Use to crawl a built static-site output dir and flag SEO/UX issues (titles, meta descriptions, alt text, thin/orphan/duplicate pages) before publishing. Triggers "audit the site", "SEO check", "check the build for SEO". Generic to any static site whose output is plain HTML.

Use when adding or fixing a mermaid diagram in a Hugo (or other static-site) page so it renders readably inline in a narrow content column (triggers "add a diagram", "mermaid", "diagram is too wide"). Covers the width rule, layout direction, the semantic color palette, and the browser viewBox check.

Use when triaging or preparing a GitHub repository security advisory as a maintainer (triggers "draft the advisory", "prepare GHSA content", "request CVE", "publish advisory"). Lists/triages advisories via the API and produces paste-ready GHSA form content; CVE-request and Publish are UI actions. Generic to any maintainer-owned repo.

Bump the pinned CLI tool versions that GitHub Actions workflows download at runtime (helm, kind, skaffold, cosign, golangci-lint, goreleaser, etc.) — the `*_VERSION:` env vars and the `# vX.Y.Z` comments next to SHA-pinned `uses:` actions. Use for "update workflow tool versions" (also invoked as `workflow-tool-versions`), "CI tool refresh", "bump skaffold/kind/cosign", or processing a Dependabot github-actions PR. Generic to any repo with GitHub Actions workflows.

Use when bumping Hugo, Go, or a theme loaded as a Hugo Module via go.mod, with versions pinned in a deploy config (netlify.toml or a GitHub Actions workflow). Triggers "update go version", "bump the theme", "upgrade hugo".

Triage and root-cause a failing GitHub Actions CI run on a PR efficiently. Use when CI is red on an open PR, a check needs triaging, the user asks "why is X failing in CI" / "a test failed, can you check", or after a push to confirm a failure is real (also reached by the older name `debug-github-ci`). Optimised for separating real regressions from pre-existing noise and getting the failing line with the cheapest log fetch first. Pairs with watch-ci for the push→monitor loop.

Use to generate branded 1200x630 social-share (OG/Twitter) card images for a site's pages, with title/tags/brand overlaid by Pillow over an AI, image, or gradient background. Triggers "make an OG image", "social card", "regenerate the share image". Generic to any Hugo-style content tree.

Run a grouped, bisectable Go dependency security sweep. Use when the user asks to upgrade outdated/vulnerable Go deps, run a dep security pass (also invoked as `go-deps-security-upgrade`), or process govulncheck/Dependabot findings. Lands one commit per logical dependency group on a dedicated branch so any regression is attributable and revertable. Generic to any Go module.

Use when the user wants to turn past Claude Code work into reusable automation — invoked as /harvest-automation, optionally with a window like "7d". Triggers include end-of-session review, "what should we automate", "mine my recent sessions", recurring manual workflows, repeated corrections, repeated permission prompts, or preferences worth capturing as a skill, CLAUDE.md entry, or memory.

Use when raising test coverage on a Go project that reports to Codecov (triggers "improve code coverage", "cover package X", "find coverage gaps"). Fetches Codecov totals, ranks low-covered packages, writes targeted tests, and verifies the delta. Generic to any Go + Codecov project.

Use when adding or committing an SVG asset (logos, icons) to keep it small — triggers "add this logo", "optimize svg", "svg is too big". Runs svgo and keeps the result only if it actually shrank.

Use when fixing or triaging GitHub code-scanning / CodeQL alerts (triggers "fix codeql issues", "check code-scanning alerts", "dismiss false-positive alert"). Lists alerts, finds the taint source, fixes real ones, dismisses won't-fix, and verifies on the PR merge ref. Generic to any repo with CodeQL enabled.

Use to pull a GA4 + Google Search Console report (top pages, queries, CTR, near-miss positions) into a dated markdown/JSON summary for an SEO/reachability pass. Triggers "analytics report", "what's my search traffic", "GSC report". Generic to any site with GA4 + Search Console configured.

Use when a PR has bot/Copilot review comments to clear — fix them, mark the threads resolved, and re-request the bot until the PR is at a good base (triggers "resolve copilot comments", "mark review threads resolved", "re-request copilot review"). Generic to any GitHub repo with bot reviewers.

Use when the user wants to obtain, inspect, or reproduce the vulnerable source code referenced by a GitHub Security Advisory (GHSA-xxxx / CVE) — including security research, PoC reproduction, patch diffing, or auditing the affected file/function.

Use when scrubbing or triaging a GitHub repo's open issue/PR backlog — e.g. "go through all open issues and PRs and see what can be closed", "scrub the outstanding issues", "close stale/duplicate items" — closing stale/duplicate/already-shipped/EOL items and categorizing the rest (type, area, priority) like a product manager. Use for one-off backlog cleanups or a recurring triage cadence. Backed by gitcrawl (local SQLite mirror, no API-quota burn). Portable across OSS repos via --repo + a per-repo config.

Use when verifying a Hugo site build before declaring it done or pushing (triggers "does it build", "verify the site", after editing layouts/SCSS/content). Builds the way the host runs it and explains why hugo --gc --quiet is not enough.

After pushing to a PR, watch its CI checks to terminal state and surface each transition as a notification instead of busy-polling. Use when you've just pushed a fix and want to know the moment a check goes green or red, or whenever the user says "watch CI", "wait for the checks", "is CI green yet", "loop on CI". Pairs with debug-ci (hand back to it on a red check) and resolve-bot-review-threads (the fix→push→resolve→re-request→watch loop).

Use when authoring or editing a blog post in a Hugo site (any theme) — triggers "write a blog post", "publish a tutorial", "add a post". Covers file layout (single file vs page bundle), front matter, the featured-image to card+OG flow, and the layout that actually renders a post.