Drop the dedicated-repo publish path. Restore GitHub Actions Pages
deploy on Agent-Skills-for-Context-Engineering while keeping the
on-page Long-Horizon Prompting brand.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
Rebrand the project site identity and public Pages path to
long-horizon-prompting. Publish via a dedicated Pages repo so the
URL matches the project instead of Agent-Skills-for-Context-Engineering.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
Surface author credit on the overview byline and shared page footers,
linking to muratcankoylan.com and x.com/koylanai.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
Add deterministic HTML/metadata/link/reference/sitemap validation, exact rubric drift checks against the authoritative skill template, generated-file drift checks including untracked artifacts, and Puppeteer production tests under the real GitHub Pages base path. Test five pages across four viewport widths, clipboard controls, keyboard tabs, full prompt visibility, downloads, and no-JS fallbacks. Pin Pages workflow actions, split read-only build from privileged deployment, pin the skill install to a full commit, and commit the npm lockfile.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
Reframe the project around the skill rather than the CDC case: add static Overview, Guide, Prompt Lab, References, and 404 pages; a copyable rubric-complete brief template; prompt/harness boundaries; gotchas and workflow; 33 caveated research/vendor references; eight generated prompt downloads; shared navigation and copy controls; accessible tab semantics; responsive editorial styling; and updated release screenshots. The CDC run is now explicitly one unreviewed case study, not the site's foundation.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
Turn the static comparison into a release site: full-height prompts on the website, capture-only bounded panes for PNGs, copy controls for every prompt and the Cursor install command, direct GitHub skill/repository/provenance links, a concise technical release explanation, responsive layout, social metadata, and an Actions-based Pages deployment. GitHub Pages still needs the one-time repository setting switched to GitHub Actions after merge.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
Collapse soft-wrapped source newlines into flowing paragraphs for display, keep section headers and lists intact, widen the lab layout, and recapture under shot-* paths so stale prematurely wrapped previews cannot linger.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
Rewrite all four right-side artifacts as copy-ready launch prompts generated with the long-horizon-prompting method. Remove every placeholder and add concrete success predicates, named non-solutions, evidence contracts, domain-specific audits, return gates, and harness boundaries. Fresh reviewers closed evaluation-leakage, production-signature, theorem-model, and security-scope defects. The UI now presents readable fixed-height prompt panes with word counts and Copy actions. Rename screenshots to optimized-* paths to prevent stale prompt previews.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
Move every redesigned PNG to a new editorial-* path, remove the superseded screenshot paths, update the README preview, and make the capture script emit only the cache-safe names. This prevents GitHub and Cursor from serving the old navy dashboard by URL.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
Replace the dark gradient dashboard, neon score cards, pill labels, and rounded panels with a restrained warm-paper system: serif display typography, black rules, generous whitespace, one muted accent, document-like prompt columns, and a methods-table scorecard. Simplify labels and recapture all 13 PNGs. The new assets are also about 1.2 MB smaller.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
Lab README covers the four use cases, the reproduce steps (fetch -> verify ->
install -> build -> capture), how to read the scorecard, and an explicit
honesty/limits section (skill's own rubric, per-pair residual risk, unreviewed
CDC proof). Add the example to the root README examples table.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
Generated by scripts/capture.mjs. Money-shot crops (hero, *-split, *-scorecard)
rendered at 2x for legible prompt text; *-full overview pages at 1x to keep the
committed set light (~6MB total).
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
build_lab.py is the single source of truth: it defines the skill's 10-dimension
pre-launch rubric and four before/after prompt pairs (ML model-improvement run,
parallel approximation-ratio proof, distributed-systems RCA, security audit),
scores both versions, and emits data/prompt-pairs.json and ui/data.js.
The 'before' prompts are competent prompt-engineered launch prompts (persona,
context, chain-of-thought, explicit format, persistence); the 'after' briefs are
the skill's pseudo-formal task brief applied to each. Each pair records a residual
risk the brief cannot remove, and the UI carries an explicit honesty note that the
rubric is the skill's own. Aggregate rubric score moves 11% -> 100% across the four.
The UI is dependency-free and file:// safe (loads window.PROMPT_LAB_DATA from
data.js); capture.mjs drives system Chrome via puppeteer-core to render it.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
.cursor/ is gitignored (project-local IDE state), so install_skill.sh is the
committed record of the Cursor install: it copies the skill directory into
.cursor/skills/ and validates the installed copy with the reference agentskills
CLI. Verified locally with skills-ref 0.1.1 (Valid skill).
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
Fetch OpenAI's published GPT-5.6 Sol Ultra Cycle Double Cover prompt and
candidate proof (the exemplar the long-horizon-prompting skill is anchored on)
into examples/long-horizon-prompt-lab/report/, with SHA-256, source URLs, and
caveats recorded in PROVENANCE.md.
verify_report.py re-extracts the prompt and asserts every block-quote in the
skill's annotated CDC reference appears verbatim in the published source (0 drift
at retrieval), so the skill cannot silently diverge from the primary artifact.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
Normalize every SKILL.md description to double-quoted YAML so the corpus,
templates, and examples have one visible convention. Harden frontmatter
parsing against partial fence matches and non-string name/description values,
and extend tests to cover those regressions.
Fix stale or misleading developer-facing docs: remove phantom book-sft plugin
install instructions, document directory-copy installs, add .agents/skills
manual install guidance, fix the digital-brain setup script target, and mark
historical 13-skill analysis as historical. Cut the prepared release metadata
to 2.3.1 across manifests and root SKILL.md.
Tighten production gates by sharing dev dependencies, making orchestrators use
strict repo validation, aligning skill-health required sections with repo
validation, and linting documented plugin install commands against the actual
marketplace manifest.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
* fix: YAML-safe skill frontmatter for cross-platform discovery
Eleven skills used unquoted descriptions containing colons, which strict
YAML parsers reject during skill discovery on Cursor, Claude Code, and
Codex. Normalize all skill frontmatter to valid YAML, add a shared
skill_frontmatter parser with PyYAML validation, fix install docs to use
directory layout, and add .agents/skills symlink plus Open Plugins
skills path declaration.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
* test: add frontmatter parser unit tests and harden examples
Add researcher/scripts/tests/test_skill_frontmatter.py (19 tests) covering
parser edge cases, a strict-YAML regression guard for the unquoted-colon
bug, format/parse round-trips, and corpus + example integration checks.
Wire unit tests into CI ahead of the strict repo gate. Quote the
description fields in the two example skills that had the same
unquoted-colon YAML hazard. Simplify CRLF handling in split_frontmatter
and document CONTRIBUTING test commands.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
* chore: harden production plugin packaging validation
Remove the committed .agents/skills symlink because symlinks are fragile
on Windows checkouts and in plugin packaging. Keep Cursor and Codex
support through Open Plugins discovery and documented directory-copy
manual installs.
Extend validate_repo.py so strict validation checks that the Open
Plugins manifest and Claude marketplace manifest name the same bundled
plugin, and that Open Plugins skills discovery resolves to the same 15
published skills as the repository.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
* ci: certify cross-platform skill compatibility
Add validate_platform_compat.py to check the production surfaces defined by
Agent Skills, Cursor, Claude Code, Codex, and Open Plugins documentation.
The gate validates published skills, manifest discovery parity, and
platform-local directory-copy installs for .cursor/skills, .claude/skills,
.codex/skills, and .agents/skills.
Install skills-ref in CI and require the upstream agentskills reference
validator in the compatibility gate. Tighten manifest path checks so both
Claude marketplace and Open Plugins skill paths must start with './', and
update operator docs to run the new gate before claiming production
readiness.
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Muratcan Koylan <muratcankoylan@users.noreply.github.com>
Ships a file-based research-to-skill operating system on top of the v2.1
skill corpus. The corpus now compounds across runs: mechanisms have a
gated promotion path, claims have explicit provenance, runs have an
enforceable state machine, skill activation is regression-tested, and the
harness is benchmarked against adversarial scenarios. A continuous loop
runs the whole thing unattended from launchd, retrieving sources over
stdlib HTTP and parking anything that needs human review.
Highlights
- Mechanism registry + accepted/rejected ledgers (5 seeded mechanisms)
- Claim provenance for 6 volatile claims across 6 skills
- Corpus index mapping skills to scenarios, mechanisms, and claims
- Activation regression fixtures (8 cases, all passing)
- Adversarial benchmark harness (7 scenarios) with append-only history
- Run state machine: initialized -> retrieved -> evaluated -> proposed ->
novelty_checked -> validated -> pr_ready -> closed
- Continuous loop (loop_discover, loop_step, loop_daily, loop_status)
with atomic JSONL writes and fcntl locks
- launchd service definitions and operator runbook
- CI workflow runs validate_repo --strict, run_benchmarks, and
check_activation_cases on every PR
- AGENTS.md, CHANGELOG.md, researcher/runs/README.md,
researcher/insights/auto-research-experiment.md
Skill changes
- All 14 published skills migrated from keyword triggers to
activation-scenario descriptions
- Hardened evaluation/advanced-evaluation and project-development/
harness-engineering boundaries
Repository policy
- Active research runs (researcher/runs/*/) are runtime state and
gitignored. Seed run kept as a worked example, closed reference-only.
- Runtime queue, logs, snapshots, status, and loop event files are
gitignored.
Verification
- validate_repo --strict: 0 errors, 0 warnings, 14 skills
- run_benchmarks: 3 checks, 0 failures, 7 scenarios
- check_activation_cases: 8 cases, 0 failures
- Python compile: clean (12 researcher scripts)
- Lint: clean
- Live loop: 13 iterations executed across init/retrieve/park/reap paths
No paid LLM judges, no automated source discovery beyond the manual seed,
and no log rotation. See researcher/insights/auto-research-experiment.md
for the full lessons-learned writeup.
Co-authored-by: Cursor <cursoragent@cursor.com>
- Replace marketplace.json with proper plugin.json manifest
- Update all SKILL.md descriptions to use third-person trigger phrases
- Add version field to skill frontmatter
- Add PLUGIN-README.md with installation docs
- Remove empty digital-brain-skill directory
Based on official structure from anthropics/claude-plugins-official
- Remove individual .claude-plugin/plugin.json from each skill directory
- Update marketplace.json to use skills array (bundle format)
- Update README with correct plugin installation commands
Expands marketplace.json to list all skills explicitly and adds plugin.json manifests to each skill directory to enable proper discovery in Claude Code.
Expands marketplace.json to list all skills explicitly and adds plugin.json manifests to each skill directory to enable proper discovery in Claude Code.
- Personal operating system for founders/creators as Claude Code skill
- 6 isolated modules following context engineering principles
- 4 automation scripts with tool-design consolidation
- Complete skills traceability documentation
- Complete pipeline for training style-transfer models from books
- Includes Gertrude Stein case study with real validation metrics
- 592 training examples, 97% loss reduction, 70% human score on Pangram
- Integrates with context engineering skills: project-development,
context-compression, multi-agent-patterns, evaluation
- Total cost: $2 for dataset generation + training
muratcankoylan