Groundledger is an MIT suite of Agent Skills for Claude Code, Cursor, and Cowork: publish-ready blogging, site SEO audits, observe-only paid media, multi-pass editing, platform-native social, and multi-surface strategy — with a hard ban on fabricated statistics.
Docs: dranshrad.github.io/groundledger · Repo: github.com/dranshrad/groundledger
Formerly published as Clearcast; renamed to Groundledger to avoid unrelated trademark/name collisions (not affiliated with Clearcast UK or other ClearCast apps).
Updated: 2026-07-25
Content and acquisition work should (1) help a reader finish a job, (2) be extractable for AI answers, (3) cast into social without inventing numbers, (4) audit sites with falsifiable recommendations, and (5) keep paid media observe-only until explicit approval. Groundledger ships CLEAR, SPARK, ORBIT, SNAP/PULSE, PROBE, SAFE, Studio Desk, and Cue Deck — plus a binding Grounding Law.
| Skill | Framework | Use for |
|---|---|---|
| groundledger | Router | Pick / compose skills |
| blog-engine | CLEAR | Articles, claim ledgers, Ship Scan, Cite Surface |
| editorial-pass | SPARK | Multi-pass polish, Tone Retarget, Voice Canon |
| social-cast | SNAP + PULSE | Hooks, posts, atomize, calendars, analytics |
| orbit-discovery | ORBIT | Surface bets, Echo Map, Close Path |
| site-signal | PROBE | Site Health Dial, page-fit, cite-AI, drift, local SEO |
| paid-cast | SAFE | Paid audits, unit econ, brand/budget lattices; MutationLatch |
| studio-desk | Ops | Workspace Boot, versions, export, Ship Gate |
| cue-deck | Prompts | Reusable grounded cue cards |
git clone https://github.com/dranshrad/groundledger.git
mkdir -p ~/.claude/skills
cp -R groundledger/skills/* ~/.claude/skills/Plugin-style (from repo root, Claude Code):
/plugin marketplace add dranshrad/groundledger
/plugin install groundledger@groundledger-marketplace
Or: claude plugin install . when this directory is the working tree (if your Claude Code build supports local plugin install).
git clone https://github.com/dranshrad/groundledger.git
mkdir -p ~/.cursor/skills
cp -R groundledger/skills/* ~/.cursor/skills/Zip each folder under skills/<name>/ (must contain SKILL.md) and upload via Cowork → Customize → Skills. Upload all nine for the full suite, or start with groundledger + blog-engine.
Project-local (either client): copy into .claude/skills/ and/or .cursor/skills/.
Details: INSTALL.md · CLAUDE.md · docs/getting-started.md.
Before any draft is “final”:
- Material claims are on a claim ledger (
verified/attributed/author-supplied) - No invented stats, studies, quotes, rankings, ROAS, or customer results
- Uncertain points use soft language or are omitted
blockedledger rows must be fixed or removed- Paid: no live spend changes unless MutationLatch is open and approved
Full rules: skills/blog-engine/references/grounding.md.
Content: orbit-discovery → blog-engine → editorial-pass → social-cast → studio-desk ship
Site SEO: site-signal audit → PROBE plan → blog-engine / orbit-discovery for fixes
Paid: paid-cast audit (observe-only) → trial / math → latch-gated optimize-draft only with approval
Three patterns here solve general agent-safety problems and transfer outside content work — copy them into any project:
| Primitive | Solves |
|---|---|
| grounding-law | Fluent, confident, unsupported claims |
| falsifiability-contract | Recommendations that cannot be checked |
| write-gate | Irreversible actions, and false completion reports |
See primitives/.
Adversarial fixtures that test the behavioural claim — bait prompts where fabricating is the path of least resistance:
python3 evals/check.py --lint # validate fixtures
python3 evals/check.py CASE.md reply.md # score a saved replySee evals/README.md. Structure and link integrity are checked by python3 scripts/validate.py (also runs in CI).
| Project | Link |
|---|---|
| Voice notes → Artifacts | voice-notes-to-anthropic-artifacts |
| LibCST LLM refactorer | llm-cst-refactorer |
| Self-correction loop | automated-self-correction-loop |
| Anthropic audio gateway | anthropic-audio-gateway |
See docs/ECOSYSTEM.md.
MIT © 2026 Divyansh Gupta