Skills
Skills
AI skills for the LTP / Issue Tree app. Kept here (top-level, not inside a runtime-specific directory) so the canonical skill can be shared by Claude Code and Codex.
- Claude Code: symlink a skill into the gitignored
.claude/skills/<name>. - Codex: symlink a skill into
.agents/skills/<name>. This repository tracks.agents/skills/ltp-project(and the deprecated.agents/skills/project-ltp) so they appear in the Codex Skills sidebar and can be invoked with$ltp-project.
Skills
| Skill | Purpose | Status |
|---|---|---|
tree-gen/ | Input claim/question → claim tree (Stream 1a) | v0.1 — verified vs gold on the toy claim, see tree-gen/VERIFICATION.md |
annotation-mapping/ | Source doc + tree → fragments mapped to nodes with relation labels (Stream 1b) | v0.1 — verified vs gold annotations (5/5), see annotation-mapping/VERIFICATION.md |
ltp-project/ | Document, repository, plan, issue export or notes → a Reason Commons *.ltp.yaml project file (the interchange document the app imports, and exports as JSON or YAML with history), plus a report of every judgment call | unified 2026-09-06 — mirror; the canonical copy is Promise-Foundation/reason-commons .claude/skills/ltp-project/, keep the two byte-identical (diff -r) |
project-ltp/ | Deprecated 2026-09-06 — its ltp-model.yaml format cannot be imported into Reason Commons. Kept only for the read-only dashboard over the legacy models still bundled on the site, and its build scripts | superseded by ltp-project/; GitHub issue sync is now the Reason Commons projection's job |
scrollable-explainer/ | Principles + patterns for writing and building long-form scroll-driven visual explainers, with a measured teardown of the ProPublica CCS piece as the worked case study | v0.1 — reference only (no code); read before drafting in ../explainers/ |
contribution-proposals/ | A batch of raw participant contributions (Google Doc export, form dump, pasted notes) + a target ltp-model.yaml → one reviewable proposal per contribution for how it might extend, support, challenge, or fail to fit the model, rendered as a self-contained page a room clicks through live — accept, edit, reject — watching the tree change. Never merges | v0.1 — renderer verified end to end in a browser against the real 2R model; the reading is untested on contributions it did not write, see contribution-proposals/VERIFICATION.md |
goal-alignment/ | Two LTP models (an individual/smaller-group one + a collective one) → reviewable, human-confirmed suggestions for where the first could advance the second — never auto-merged | pilot #1 — toy example only, see goal-alignment/VERIFICATION.md; the IO-5 AI-vs-steward comparison it enables has not yet run |
The canonical tree JSON schema is defined in tree-gen/SKILL.md; tree-gen/gold/remote-work.tree.json is the gold fixture (extracted from ../claim-tree-annotation.md) that also seeds 1b and 1c.
Each is validated against the gold-standard toy case in ../claim-tree-annotation.md (remote-work productivity). See each skill's own VERIFICATION.md for its pass criteria.
