A guide-refresh + ecosystem catch-up minor release shipped in five passes against a research-grounded plan. No breaking changes.
Inventory delta (v1.8.0 → v1.9.0)
| v1.8.0 | v1.9.0 | Delta | |
|---|---|---|---|
| Skills | 30 | 36 | +6 |
| Agents | 14 | 16 | +2 |
| Rules | 24 | 26 | +2 |
| Hooks | 6 | 6 | 0 |
New capability
New skills (6):
- `/humanize` — detect-and-flag for AI-voice tells in academic prose (read-only; no rewrite)
- `/prompt` — informal/dictated input → structured six-section prompt + execute (ported from Blattman v2.1)
- `/prompt-only` — same formatting, no execution (reusable prompt artifact)
- `/compress-session` — distil long sessions into structured notes before auto-compaction
- `/promote-memory` — five-critic council (generality/staleness/redundancy/evidence/format) for `[LEARN]` promotion to MEMORY.md
- `/stata-replication` — end-to-end Stata pipeline via the `stata-mcp` MCP server
New agents (2):
- `humanize-auditor` — read-only AI-voice auditor
- `promote-memory-council` — five-critic implementation
New rules (2):
- `model-routing.md` — 70/20/10 architect/editor split (Haiku/Sonnet/Opus)
- `stata-code-conventions.md` — header scaffold, numbered pipeline, esttab tables, AEA compliance
New template: `passport-template.yaml` (per-paper claims provenance with PASS/FAIL/STALE/UNVERIFIED status)
New reference: `prompt-formatting-core.md` (shared by `/prompt` and `/prompt-only`)
Other capability
- `/review-paper --variance N` — reviewer-disposition variance mode. Returns a decision distribution (not a point estimate). Motivated by AgentReview (ACL 2024).
- HIGH-WARN gate-refuse on `/commit` — fabricated citations / numerical contradictions block the commit unless explicit override.
- Pattern 16: Preregistration and Submission Discipline — full workflow + per-registry mandatory-field matrix (OSF / AsPredicted / AEA RCT).
- Apr–May 2026 Anthropic catch-up — Opus 4.7, `/goal`, `claude agents`, `worktree.baseRef`, `xhigh` effort level, `/fewer-permission-prompts` typo fix, auto-mode flag retirement.
- Cost-Conscious Composition — prompt-cache TTL change, 70/20/10 model routing, `/cost` / `/usage`, Agent SDK credit-pool split (2026-06-15).
- Ecosystem updates — clo-author v26.05, stata-mcp.
- Apr 23 2026 Anthropic "Quality reports update" cited in verification-discipline section as evidence that model quality drifts.
PRs
#114 (Pass 1) · #115 (Pass 2A) · #116 (Pass 2B) · #117 (Pass 2C) · #118 (Pass 2D) · #119 (Pass 3A) · #120 (Pass 3B) · #121 (Pass 4) · #122 (Pass 5)
Verification (every PR)
- `./scripts/check-surface-sync.sh` — 26/26 assertions pass
- `./scripts/check-palette-sync.sh` — in sync
- `check-skill-integrity` — all checks pass
- `quarto render guide/workflow-guide.qmd` — clean
- `python3 scripts/quality_score.py guide/workflow-guide.qmd` — 100/100 [EXCELLENCE]
🤖 Generated with Claude Code