Skip to content

feat(core,ci,docs): built-in skills (15) + effort-bench + release pipeline + BEHAVIOR_PARITY#14

Merged
oratis merged 1 commit into
mainfrom
feat/builtin-skills-effort-bench
May 28, 2026
Merged

feat(core,ci,docs): built-in skills (15) + effort-bench + release pipeline + BEHAVIOR_PARITY#14
oratis merged 1 commit into
mainfrom
feat/builtin-skills-effort-bench

Conversation

@oratis

@oratis oratis commented May 28, 2026

Copy link
Copy Markdown
Owner

Summary

4 bundled deliverables in one PR:

  1. 15 built-in skills as SKILL.md files in packages/core/skills/ — bundled in the npm package, auto-loaded by CLI alongside user/project layers
  2. scripts/effort-bench.ts — runs 3 scenarios × 5 effort tiers against real DeepSeek API, writes measured numbers to docs/design/effort-levels-measured.csv
  3. .github/workflows/release.yml — tag-triggered release pipeline (npm publish + GitHub release). Mac client step stubbed if: false until M6 lands Electron app
  4. docs/BEHAVIOR_PARITY.md — comprehensive per-feature comparison with Claude Code

Release notes

  • release-notes:feature

Co-Authored-By: Claude Opus 4.7 (1M context) noreply@anthropic.com

… + behavior parity

What ships
----------
Built-in skills (15 SKILL.md files in packages/core/skills/)
- init / verify / run / code-review / security-review / skill-creator
- consolidate-memory / fewer-permission-prompts / update-config / keybindings-help
- deepseek-api / loop / schedule / review / pdf
Each has frontmatter (name + description) and a short body. The CLI auto-locates
this dir via `@deepcode/core/package.json` resolution and passes it to loadSkills
as the `builtinDir`. The 15 are baseline stubs — content can grow over time.

Effort benchmark (packages/core/scripts/effort-bench.ts)
- 3 scenarios × 5 effort tiers = 15 real API calls
- Reads creds from DEEPSEEK_API_KEY env OR ~/.deepcode/credentials.json
- Writes docs/design/effort-levels-measured.csv next to the design doc
- Computes ¥ cost using DeepSeek pricing from docs/design/effort-levels.md §2.4
- Run: pnpm -F @deepcode/core tsx scripts/effort-bench.ts

Release pipeline (.github/workflows/release.yml)
- Triggered on git tag v*
- validate job: typecheck + test + build on Ubuntu Node 22
- publish-cli: npm publish, with --tag <channel> for beta/nightly
- build-mac: stubbed (if: false) until M6 ships Electron app — has the full
  electron-builder + Apple notarization + artifact upload structure ready
- github-release: auto release notes from `git log`, prerelease flag for non-stable
- Channel/version/mandatory derived from tag: vX.Y.Z (stable), vX.Y.Z-beta.N,
  vX.Y.Z-nightly.N, vX.Y.Z+security.N

BEHAVIOR_PARITY.md (docs/)
- Per-feature comparison table vs Claude Code: 30+ slash commands, hook events,
  hook handler types, modes, memory system, MCP, tools, CLI flags
- Each row marked ✅ matches / 🟡 caveats / 🔄 deferred / ⚠️ deliberately differs / 🆕 DeepCode-only

REPL plumbing (apps/cli/src/repl.ts)
- New `resolveBuiltinSkillsDir()` helper finds packaged skills/ via
  @deepcode/core/package.json + dirname; loaded into skills registry alongside
  user/project/plugin layers

Package manifest
- packages/core/package.json `files` now includes "skills/" so npm install
  bundles the built-in skills

Verified
--------
  pnpm typecheck   → green
  pnpm test        → 308 passed / 8 skipped / 0 failed (unchanged — skills are content)
  pnpm format:check → conformant

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@oratis oratis merged commit e91aaed into main May 28, 2026
@oratis oratis deleted the feat/builtin-skills-effort-bench branch May 28, 2026 04:46
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant