Skip to article frontmatterSkip to article content
Site not loading correctly?

This may be due to an incorrect BASE_URL configuration. See the MyST Documentation for reference.

Future skill ideas

Documented so they aren’t lost; deliberately not committed to. What has actually shipped is in Skill catalog, and work in flight is in the per-plugin tracking issues. Each idea below is evidence-backed by the 2026-07 PR-history analysis (~630 merged PRs across lecture-python.myst, lecture-python-intro, lecture-python-advanced.myst, lecture-dp).

Parked candidates

mirror-lecture (cross-repo SYNC)

Mirror an upstream lecture change into a downstream repo. 13 of lecture-dp’s 28 lifetime PRs are SYNC: PRs; the same infra fix (Pages deploy migration, linkcheck fix, fetch-depth) has landed 2–4× across repos by hand. The playbook is stable and documented in existing PR bodies: pin upstream commit hash, localize {doc} cross-references, defer TOC wiring, neutral commit-reference wording (no closing keywords before owner/repo#N).

triage-build (environment-upgrade breakage)

Classify a failed lecture build against known failure patterns (deprecation sweeps, pandas 3.0 read-only arrays, unpinned-package breakage à la prettytable, SymPy overflow) and propose the known fix. Recurs ~15×/year across repos. Overlaps meta#304’s jupyter-book-fixer idea.

migrate-rng (np.random → Generator campaign)

meta#299: ~13 identical PRs so far, many lectures remaining. Mechanical with a defined judgement checkpoint (figure changes from RNG-stream changes). Time-limited — only worth a skill if the campaign runs long. The unsettled convention (rng explicit arg vs rng=None fallback — lp #874 vs intro #741) should be settled in the manual regardless.

intake-lecture (LaTeX → MyST intake)

“Tom’s lectures” intake formatting: ~20 PRs in lp + adv converting delivered material to house MyST style. Maps to meta#304’s latex-to-myst. Real but bursty.

migrate-data (data-lectures repointing)

meta#338 / data-lectures#15 playbook: data PR first, byte-compare sha256 gate, repoint URLs, delete local copies (lp #973, intro #792). Pure skill material, low current volume.

Editorial-round implementation

Parse a review issue’s checklist into a diff against one lecture — lecture-python-intro’s dominant RA workflow (~30 PRs). Largely absorbed by the style skill’s fix mode; revisit only if a gap remains after it ships.

A naming convention, once there is something to generalise from

Two shapes are in use — the verb in the skill (/qe:check-style) and the verb in the plugin (/audit:issues) — and developing-skills.md deliberately declines to rank them. Worth revisiting once enough plugins exist to show which reads better in practice, and specifically whether mixing both shapes inside one plugin actually causes trouble or only looks untidy. Writing the rule now would mean generalising from three plugins and no usage.

There is now one instance to watch: qe holds both /qe:check-style (verb first) and /qe:copilot-review, which is a noun phrase — the object is the Copilot review, and the verb is only implied. It was named for what a user would type when they mean “deal with the Copilot review”, not to fit either shape. Whether the mismatch costs anything in the slash menu is the observation to collect.

How skills could serve the benchmarking programme (note for meta#335)

Benchmark data capture stays in the programme repos (QuantEcon/benchmarks — workstream A; QuantEcon/tool-lecture-benchmark — workstream C). Skills are natural consumers and interfaces of that data:

  1. .jupyter_cache telemetry reader — nbclient already records per-cell timings in every CI build; nothing surfaces them. A skill + ~50-line script reading _build/.jupyter_cache (or a CI artifact) and diffing a PR preview against main (“cell solve-ex4 went 8 ms → 950 ms”) is the cheapest high-value entry point and seeds workstream C. It would have flagged the lecture-python.myst#717 regression the day the PR opened.

  2. Crossover advisory — once workstream A produces crossover curves, a skill answering “this lecture solves n=2 economies 100 times — which backend?” from measured data rather than rules of thumb, during authoring or review.

  3. Provenance/schema guardian — generate and validate the shared result + environment-descriptor stamp (meta#335’s first blocking deliverable) across all three workstreams’ outputs.

  4. /benchmark:review-acceleration is workstream B’s deliverable and already lives here.

Rejected

Project
Skill catalog