Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Standardize how an AI-collaboration project turns work into reusable knowledge. Use when initializing or retrofitting Project Cairn in a project, recording progress after meaningful work, maintaining AGENTS/CLAUDE/cairn docs, auditing project knowledge for drift or missing records, pulling and citing external knowledge, or graduating validated project experience into a reusable knowledge base.
.claude/skills/iblinkq-project-cairn/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 28% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -2% | 0% |
Project Cairn gives AI-collaboration projects a durable way to keep and reuse what they learn while doing real work. The Skill supplies the method, AGENTS.md carries always-read project rules, and .cairn/config.yaml stores machine-readable configuration.
This Skill is documentation-first. It provides instructions, templates, references, and provider adapters. It has no CLI, MCP server, background hook, chat-end trigger, automatic provider write, or automatic history migration. Routine maintenance happens during normal work because AGENTS.md carries the rules. Provider writes require human confirmation.
Load exactly the reference you need:
references/init.md — initialize, retrofit, bootstrap, or set up Project Cairn in a project.references/consume.md — pull, cite, reuse, or apply external knowledge from the configured knowledge base.references/maintenance.md — record progress, update LOG, update ROADMAP, maintain topic notes, or apply Cairn rules after work.references/audit.md — inspect, audit, validate, clean up, or find missing project knowledge records.references/upgrade.md — upgrade an initialized project instance to the current skill spec, or check how far it has drifted (instance drift after the skill itself evolved).references/graduation.md — graduate knowledge, move reusable experience to a knowledge base, or judge graduation-readiness; a provider write additionally reads only the selected references/graduation/<provider>.md.references/frontmatter.md — create or review Cairn topic notes or knowledge-base notes.references/branch-closure.md — review, close, abandon, or roll back an exploration branch and salvage its cairn/ knowledge before it disappears.references/zh-glossary.md — writing project docs in Chinese (or resolving what a Chinese term should be); fixed English↔Chinese terminology mapping for Project Cairn's own vocabulary.Frontmatter describes triggers, not workflow.
assets/templates/ contains AGENTS.md, CLAUDE.md, config.yaml, user-config.yaml, LOG.md, ROADMAP.md, topic.md, and Cited.md. init scaffolds AGENTS.md, CLAUDE.md, .cairn/config.yaml, cairn/LOG.md, and optional cairn/ROADMAP.md; it can seed optional ~/.config/cairn/config.yaml from user-config.yaml. topic.md, Cited.md, and Reference/ are trigger-created. Templates use {{PLACEHOLDER}} tokens; never hardcode a provider or knowledge path.
AGENTS.md is read as project rules.scripts/*.sh use bash (verified on macOS/Linux; Windows needs WSL or Git Bash). scripts/*.py use Python 3 without a shell; notion-graduate-batch.py also needs PyYAML.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 15,921 | 19,602 | +23% | 1 | 1 | 0% | 2,924 | 3,831 | +31% | 0 | 0 | — |
case-02 | fail→pass | 16,968 | 12,591 | -26% | 1 | 1 | 0% | 2,548 | 2,736 | +7% | 0 | 0 | — |
case-03 | fail→fail | 5,141 | 7,687 | +50% | 1 | 1 | 0% | 670 | 1,916 | +186% | 0 | 0 | — |
case-04 | fail→pass | 11,743 | 10,268 | -13% | 1 | 1 | 0% | 1,856 | 2,367 | +28% | 0 | 0 | — |
case-05 | fail→pass | 12,469 | 3,346 | -73% | 1 | 1 | 0% | 2,009 | 1,194 | -41% | 0 | 0 | — |
case-06 | fail→pass | 12,713 | 7,808 | -39% | 1 | 1 | 0% | 2,050 | 2,010 | -2% | 0 | 0 | — |
case-07 | fail→pass | 22,502 | 11,869 | -47% | 1 | 1 | 0% | 3,553 | 2,461 | -31% | 0 | 0 | — |
case-08 | fail→pass | 15,733 | 12,164 | -23% | 1 | 1 | 0% | 2,339 | 2,514 | +7% | 0 | 0 | — |
case-09 | fail→pass | 12,008 | 2,984 | -75% | 1 | 1 | 0% | 1,557 | 1,187 | -24% | 0 | 0 | — |
case-10 | fail→pass | 13,637 | 3,762 | -72% | 1 | 1 | 0% | 2,145 | 1,296 | -40% | 0 | 0 | — |
case-11 | fail→pass | 15,816 | 11,989 | -24% | 1 | 1 | 0% | 2,272 | 3,005 | +32% | 0 | 0 | — |
case-12 | fail→pass | 11,563 | 4,028 | -65% | 1 | 1 | 0% | 1,675 | 1,267 | -24% | 0 | 0 | — |
case-13 | fail→pass | 9,990 | 4,071 | -59% | 1 | 1 | 0% | 1,570 | 1,260 | -20% | 0 | 0 | — |
case-14 | fail→pass | 10,190 | 3,951 | -61% | 1 | 1 | 0% | 1,487 | 1,203 | -19% | 0 | 0 | — |
case-15 | fail→pass | 5,458 | 2,922 | -46% | 1 | 1 | 0% | 879 | 1,082 | +23% | 0 | 0 | — |
case-16 | pass→pass | 12,019 | 6,721 | -44% | 1 | 1 | 0% | 1,833 | 1,561 | -15% | 0 | 0 | — |
case-17 | pass→pass | 8,810 | 4,308 | -51% | 1 | 1 | 0% | 1,214 | 1,233 | +2% | 0 | 0 | — |
case-18 | pass→pass | 9,128 | 4,311 | -53% | 1 | 1 | 0% | 1,234 | 1,199 | -3% | 0 | 0 | — |
case-19 | pass→pass | 17,160 | 9,443 | -45% | 1 | 1 | 0% | 2,152 | 1,944 | -10% | 0 | 0 | — |
case-20 | pass→pass | 11,149 | 9,935 | -11% | 1 | 1 | 0% | 1,553 | 2,045 | +32% | 0 | 0 | — |
case-21 | fail→pass | 7,528 | 6,566 | -13% | 1 | 1 | 0% | 1,068 | 1,530 | +43% | 0 | 0 | — |
case-22 | pass→pass | 11,444 | 9,894 | -14% | 1 | 1 | 0% | 1,875 | 2,585 | +38% | 0 | 0 | — |
case-23 | pass→pass | 12,000 | 11,338 | -6% | 1 | 1 | 0% | 2,065 | 2,565 | +24% | 0 | 0 | — |
case-24 | pass→pass | 5,864 | 6,540 | +12% | 1 | 1 | 0% | 772 | 1,620 | +110% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 24 cases were attempted. The headline lift of +63 percentage points is the difference between those two pass rates over the 24 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.