Skip to content

v2.0-backlog.md

Latest commit

 

History

History
112 lines (72 loc) · 14.4 KB

v2.0-backlog.md

File metadata and controls

112 lines (72 loc) · 14.4 KB
Raw

v2.0 Backlog

Forward-looking work explicitly deferred from earlier releases. Lives here (committed) so forkers can see roadmap signals without reading gitignored plan files. Updated as items land or get re-prioritized.

Last reviewed: 2026-06-09 (post-v2.0.0 ship) Previous file: v1.9-backlog.md — renamed to v2.0-backlog.md on 2026-05-20 after v1.9.0 shipped; CHANGELOG references to the old filename are historical.

Shipped in v1.9.0 (no longer in backlog)

The v1.9.0 release closed the following items that were on the v1.9 backlog at v1.8.0 ship:

  • ✅ Capability skills — none of /syllabus, /teach-from-paper, /power-analysis, /replication-package, /respond-to-eval shipped this cycle (these remain deferred — see below). However, 6 new skills outside the original v1.9 capability list shipped: /humanize, /prompt, /prompt-only, /compress-session, /promote-memory, /stata-replication. See CHANGELOG.md v1.9.0 for full details.
  • ✅ Personal-style-guide reference (partial) — was originally framed as a voice extraction template. Adjacent capability shipped: /humanize (detect AI-voice tells in academic prose). A positive voice-profile reference (extract YOUR style from prior papers, audit drafts against it) remains deferred — see below.
  • ✅ Stata expansion — /stata-replication + .claude/rules/stata-code-conventions.md + /audit-reproducibility Stata coverage shipped (Pass 4). Was not originally on the v1.9 backlog but added per user request mid-cycle.

Shipped in v1.10.0 (no longer in backlog)

The v1.10.0 "hub-expansion" release shipped the first slice of the multi-project-hub mission surfaced by the 2026-05-31 repo audit (quality_reports/plans/2026-05-31_repo-improvement-roadmap.md):

  • ✅ Monte Carlo simulation capability — /simulation-study skill + sim-reviewer agent + simulation-conventions rule. Closes the gap where simulation appeared only as something /review-paper critiqued, never something the template produced.
  • ✅ Hub framing — "One repo, many project types" section in README + landing page.
  • ✅ R package-development gate — /r-package-check skill + r-package-reviewer agent + r-package-conventions rule (plus an analysis-only scope banner on r-code-conventions.md). Closes the largest uncovered portfolio category for R-first maintainers (12+ R packages).

Shipped in v2.0.0 (no longer in backlog)

The v2.0.0 "modernization" release (2026-06-09) shipped the loop-first / gate-enforced paradigm shift plus the long-deferred capability skills. See CHANGELOG.md v2.0.0.

  • ✅ All five long-deferred capability skills — /syllabus, /teach-from-paper, /power-analysis, /replication-package, /respond-to-eval (deferred since the v1.9.0 plan) all shipped, plus /capture-environment, /did-event-study, /disclosure-check, /grant-proposal, /data-management-plan, /coauthor-brief, /scaffold-exercises, /new-skill, and /triage-inbox.
  • ✅ /triage-inbox — shipped (human-gated email/calendar triage + referee-obligations tracker). The /vault-note skill and a populated .mcp.json remain deferred (below).
  • ✅ Enforcing pre-commit hook + table-row sync gate — closes the "direct git commit bypasses the gate" hole and the enumerative-table drift this backlog flagged as high-priority.
  • ✅ Real orchestration runtime — orchestrator-protocol.md rewritten from "pattern, not a runtime" into fan-out → reduce → judge + hallucination gate → loop-until-dry (orchestration-schemas.md, agent-fleet.md).
  • ✅ Verification 2.0 (EXPLAINED), per-agent model/effort pinning, enforcing hooks (git-guardrails, claim-reconcile, auto-writing session log), output-styles, CI, and the confidential-data + prompt-shaping rules.
  • ✅ Retired /prompt, /prompt-only, and verify-reminder.py.

Still deferred (post-v2.0)

From the 2026-06-10 currency audit (do-soon tier)

  • Awesome-list distribution — after topics/homepage/CITATION.cff land: one-line PRs to hanlulong/awesome-ai-for-economists (CC0) and the main claude-code awesome list(s). Low effort, high reach for the exact audience.
  • Zenodo DOI — connect the GitHub repo to Zenodo (owner's account), then the next tagged release mints a DOI automatically; add the DOI badge to README + the doi field to CITATION.cff. The citability flywheel for replication-package acknowledgements.
  • Hook touchpoints note for orchestrator-protocol.md / scheduled-routines.md — optional wiring (reduce-trigger / failure-resilience / per-reviewer logging hook events). Verify each event name against current docs before documenting — this audit caught fabricated CLI syntax; don't repeat it with hook names.
  • Persona-segmented cost table (grad student vs faculty) — a small "what a month costs" table in the guide's Cost-Conscious Composition section; the model-routing rule now carries a one-line version.
  • Non-Claude coauthor gate note — document the path for a coauthor who pulls the repo without Claude Code: ./scripts/install-hooks.sh works for them too (pre-commit is plain bash/python), but skills/hooks guidance assumes the Claude loop; a short "for your non-Claude coauthor" README box would close it.
  • Audience scope (owner-set, 2026-06-10): econ + closely related fields. The psychology/sociology/public-health cards below remain deferred indefinitely — do not build without an explicit owner ask.

Portfolio hub: package development — Stata & Python (R shipped in v1.10.0)

The R package triad shipped in v1.10.0. Stata and Python remain, to cover the rest of the owner's package portfolio (2 Stata packages, 1 Python package).

  • stata-package-conventions.md rule + /stata-package-check skill — .ado / .sthlp structure, cscript certification, .pkg / .toc build, SSC-submission preflight. Sibling to the analysis-only stata-code-conventions.md.
  • python-package-conventions.md rule + /python-package-check skill — src layout, ruff / mypy, build + twine PyPI preflight. Lower priority than Stata.

Portfolio hub: personal productivity & knowledge vault

Also surfaced by the audit; lower priority and a scope decision (centralize here vs. keep in the dedicated outlook-attention-triage / DiD-vault repos with only an MCP bridge here).

  • /vault-note skill — capture a research insight into the knowledge vault with backlinks + tags. (/triage-inbox shipped in v2.0.)
  • sample .mcp.json — opt-in wiring for Gmail / Calendar + econ-data MCPs (FRED / World Bank / BLS) + Zotero.

À-la-carte packaging (deferred)

Considered for v1.10.0 and pulled before ship: a .claude-plugin/marketplace.json exposing review-suite / simulation-suite as installable subsets. The strict:false in-place schema has known path-resolution quirks (#11278) and needs interactive validation. Re-affirmed as a documented non-goal in v2.0 (see CHANGELOG v2.0.0 "Notes"); the template stays fork-the-whole-thing until #11278 resolves.

Disciplinary breadth (continuation of v1.8.0 + v1.9.0 work)

The v1.8.0 cycle shipped political-science as a concrete second-discipline alongside econ. The v1.9.0 cycle stayed within those two disciplines (no new fields added). Three other social-science disciplines remain deferred:

  • Psychology / experimental — flagship journals (Psychological Science, JPSP, Cognition); paper types include behavioural / lab-experiment / cognitive-modelling. Heavy preregistration culture (OSF default). discipline-cards.md skeleton ready; needs concrete journal profiles + sanity checks for replication-crisis era norms (e.g., manipulation checks, attention checks, MTurk quality screens).
  • Sociology — ASR, AJS, Social Forces, Demography. Paper types include qualitative case study, ethnographic, mixed-methods, network analysis. Methods-referee taxonomy needs new types (qualitative-case-study, ethnographic, mixed-methods).
  • Public health / epidemiology — AJPH, NEJM, Lancet. Heavy STROBE / CONSORT / RECORD adherence. Probably needs its own preregistration style (ClinicalTrials.gov / ISRCTN, distinct from /preregister's OSF/AsPredicted/AEA-RCT options).

For each: ~3 journal profiles in journal-profiles.md (~120 lines), 2-3 paper types in methods-referee.md, 1 card in discipline-cards.md. Roughly the same diff size as the poli-sci v1.8.0 work.

Theorist + theorist-critic agents (port from clo-author v4.2.0)

clo-author v4.2.0 added theorist + theorist-critic for drafting formal-theory sections (identification, consistency, asymptotic normality, proofs). Useful complement to our v1.8.0 formal-theory paper-type rubric in methods-referee. With permission + attribution, port them to .claude/agents/. Pair with /review-paper --peer to handle theory-only papers more rigorously than current methods-referee can alone.

Capability skills (all shipped in v2.0.0)

/syllabus, /teach-from-paper, /power-analysis, /replication-package, and /respond-to-eval — deferred since the v1.9.0 plan — all shipped in v2.0.0, alongside /capture-environment, /did-event-study, /disclosure-check, /grant-proposal, /data-management-plan, /coauthor-brief, and /scaffold-exercises. See the "Shipped in v2.0.0" section above and CHANGELOG.md.

Mechanical / discipline upgrades

  • Personal-style-guide reference (positive direction) — voice extraction template. User points at 2-3 prior papers; Claude extracts lexicon, sentence rhythm, intro patterns. Used by /respond-to-referees, /research-ideation rationale sections, and any draft-side work to keep voice consistent across an author's corpus. Complement to /humanize (which is the negative direction — detect AI voice). Together: detect AI tells (negative) + audit against your own profile (positive). Output: style-profile.md referenced by humanize-auditor to respect documented authorial preferences (e.g., "I use em-dashes deliberately"). On v1.9.0 backlog at ship time; remains deferred.
  • Enumerative-table consistency check — add a mechanical check (or extend check-surface-sync.py) that counts .claude/{skills,agents,rules}/* and verifies appendix table row counts match. Pet-peeve #18 documents the class. Promoted to high-priority after v1.9.0 added 6 skills + 2 agents; manual appendix-table updates were a meaningful share of every PR. A mechanical check would have caught the pre-existing v1.5.0 peer-review-trio drift in 1 second instead of 3 releases.
  • Generalize check-skill-integrity.py tool-parity check beyond Task. Hard-coded Task-only check missed Monitor in v1.8.0 (pet-peeve #19). Add a known-tool list updated whenever Anthropic ships a new tool primitive.
  • Auto mode messaging — README Quick Start currently advises bypass mode; consider promoting auto mode for forkers on qualifying plans. Per feedback_bypass_permissions.md memory, the shipped defaultMode stays bypass; this is a docs-only change. v1.9.0 noted auto mode no longer needs the --enable-auto-mode flag (now on Team / Enterprise / API and rolling out to Max; needs Opus 4.6+ or Sonnet 4.6); further normalising the messaging is the v2.0 follow-up.

New deferred items surfaced during v1.9.0

  • /promote-memory autonomous mode — currently the skill spawns 5 critics and presents verdicts for the user to approve. v2.0 could add --auto for unanimous (5-of-5 YES) promotions to skip the manual approval gate. Risk: the user's domain judgment is the documented final gate; auto-mode breaks that. Defer until we have telemetry showing how often 5-of-5 votes happen and how often the user actually approves them.
  • /compress-session --restore <path> — read a prior compression file and pre-populate the new session's context with the structured summary. Would close the loop on the distil-then-resume workflow. Mechanically simple (one Read + one structured prompt prefix); deferred for now because manual claude --resume + reading the file works.
  • /humanize --strict <threshold> — gate-refuse /commit if HIGH-severity tells per 1000 words exceed <threshold>. v1.9.0 deliberately ships advisory-only (per the open question in the v1.9.0 plan). Worth revisiting after users build judgment about what threshold is right for their writing style.
  • /prompt-refine — third variant of the Blattman port (audit/improve an existing prompt). v1.9.0 shipped only /prompt + /prompt-only. Defer until we have evidence that prompt-quality drift is a real problem in the wild (most users iterate by rewriting, not by auditing).
  • Passport claim_audit schema extension — current passport scope is numeric-claim provenance only. ARS's design also tracks citation provenance + literature-corpus state. Extending the schema to cover citation provenance would let /verify-claims write to the passport instead of producing standalone reports. Worth doing after we have telemetry on how often the passport is consulted.
  • Cost-Conscious Composition telemetry — the 70/20/10 routing pattern is documented in model-routing.md but per-agent model assignments are not all updated. Going through every agent and explicitly setting model: per the routing rule, then measuring actual cost impact, is the natural v2.0 follow-up. Until that happens, the rule is advisory.

Plan-level decisions, not yet made

  • Should quality_reports/ directly-under-root files (not in subdirectories) be tracked or gitignored? Currently the .gitignore only catches subdirectory contents — top-level files (like a hypothetical quality_reports/RELEASE-NOTES.md) would be tracked. Decide whether to expand the ignore pattern or document the convention.
  • Whether to add a per-discipline domain-reviewer prebuilt (instead of template-only). Currently .claude/agents/domain-reviewer.md is a template forkers customize for their field. v1.8.0 added an econ + poli-sci dual customization in-line. v2.0 could ship domain-reviewer-econ.md, domain-reviewer-polisci.md as concrete instances and keep the template as a starter for unsupported fields.

How to add to this backlog

When deferring an item from a release, add it here with:

  • What (one sentence)
  • Why deferred (scope, risk, dependency)
  • Cost estimate (lines / sessions / new agents)

Promote items out when they ship — move to that release's CHANGELOG entry, delete here.