Skip to content

Releases

v1.0.3

v1.0.3 Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 06 Nov 13:06
v1.0.3
d306917

Full Changelog: v1.0.2...v1.0.3

Assets 3
Loading

v1.0.2

v1.0.2

Choose a tag to compare

View all tags
@github-actions github-actions released this 06 Nov 12:57
v1.0.2
caec989

Full Changelog: v1.0.1...v1.0.2

Loading

Onramp 1.0.1

Onramp 1.0.1

Choose a tag to compare

View all tags
@github-actions github-actions released this 06 Nov 10:30
v1.0.1
34ee483

Fixed

  • scan.py skips installed agent skills (.claude/skills/, .agents/skills/ and similar), so a project-level install of Onramp or any other skill is not counted as the project's own code. Found in the end-to-end install test of 1.0.0.

Install: npx skills add SanyamWadhwa07/onramp or download onramp-skill.zip below. Website and live demos: https://onramp-skill.vercel.app

Loading

Onramp 1.0.0

Onramp 1.0.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 06 Nov 10:27
v1.0.0
828917f

Onramp turns any codebase into a personal, verified route through the code.

Install: npx skills add SanyamWadhwa07/onramp, or download onramp-skill.zip below and unzip it into your agent's skills folder.

Website and live demos: https://onramp-skill.vercel.app

First public release.

Added

  • The skill (skills/onramp/SKILL.md): full guide, quick, explain, trace, update and
    tutor modes, written for any agent that reads SKILL.md and runs Python.
  • scan.py: inventory, manifests, entry points, import graph for JS/TS, Python, Go, Rust,
    Java/Kotlin, C/C++, Ruby and PHP, importance ranking, git history, env vars, TODOs, deleted
    files and Claude Code session detection.
  • sessions.py: opt-in digest of local Claude Code transcripts with secret redaction.
  • verify.py: checks every [[path:lines|anchor]] citation against the code, re-anchors
    drifted lines, rejects unsupported claims, merges and repairs work fragments, checks single
    fragments for subagents.
  • render.py: one self-contained HTML guide with the real code embedded, plus GUIDE.md.
  • Viewer: route planner for 1 hour to 1 week and four goals, the road map with progress,
    lessons with checkpoints, hands-on stops on longer routes, code panel, file explorer, computed
    dependency map, search, light and dark themes.
  • Evals: fault injection for the verifier, FActScore-style grounding and SWE-QA
    comprehension, with every prompt, answer and grade in evals/.
  • Website with live PaperMind and nanoGPT demos, deployed on Vercel.

Evaluation: fault injection 525/525 faults caught with 0 false alarms; FActScore 96.7% of sampled claims supported; SWE-QA (Flask) 67.1/100, level with the model's own knowledge. Method and every grade: https://github.com/SanyamWadhwa07/onramp/tree/main/evals

Loading