Closed
Labels
v0.2Targeted for the v0.2 readiness milestoneTargeted for the v0.2 readiness milestone
Milestone
Description
What v0.2 means
deepsec v0.1 (released a week ago) was feature-complete on the engineering axis: full pipeline, self-improving loop, 12-language coverage, AST grammars, CI integration. v0.2 closes the gap to real users — non-internal contributors, external installs, public scrutiny.
Status (2026-05-14)
Docs (high priority — friction-blockers for adoption)
- Docs: CONTRIBUTING.md rewrite for the Go era #93 —
CONTRIBUTING.mdrewrite for the Go era (shipped in docs: CONTRIBUTING (Go era) + SECURITY + index #107) - Docs: SECURITY.md + threat model (responsible disclosure) #94 —
SECURITY.md+ threat model (shipped in docs: CONTRIBUTING (Go era) + SECURITY + index #107) - Docs: top-level index + troubleshooting #97 —
docs/index.md+docs/troubleshooting.md(shipped in Ops:deepsec doctordiagnostic command (closes #95) #106 / docs: CONTRIBUTING (Go era) + SECURITY + index #107)
Operational tooling (friction-blockers for daily use)
- Ops:
deepsec doctordiagnostic command #95 —deepsec doctordiagnostic command (shipped in Ops:deepsec doctordiagnostic command (closes #95) #106) - Ops:
deepsec spendcost aggregation across runs #96 —deepsec spendcost aggregation (shipped in ops:deepsec spendcost aggregation across runs #108)
Evidence collection — the loop running in anger
- Run deepsec on 5 popular OSS repos (Express, Flask, Rails, Next.js, jwt-go) — first batch #98 — Run deepsec on 5 popular OSS repos (shipped in scans: first batch (Express, Flask, Rails, Next.js, jwt-go) — closes #98 #100; writeups at docs/scans/2026-05-14-first-batch.md)
- Bench corpus: 50-100 labeled fixtures from CVEs / Security Advisories #84 — Bench corpus growth — 30 tasks (was 4 at start; grown across matchers: ruby-yaml-unsafe-load (CVE-2013-0156 family) #109, matchers: open-redirect-from-referrer + bench fixture #110, matchers: nextjs-draft-mode-redirect + bench fixture #111, corpus: negative fixtures from flask + jwt-go cores #112, corpus: bench tasks for Java + Rust + C (toward #84) #113, corpus: Kotlin + Swift + Java JWT fixtures (toward #84) #115, corpus: rust-axum-no-auth + java-xxe + swift-keychain (toward #84) #116, corpus: batch 4 — Java Spring CORS + Rust serde + C rand #117, corpus: batch 5 — Java SSRF + Rust transmute + C system #118, corpus: batch 6 — C TOCTOU + C memcpy + Kt WebView #119, corpus: batch 7 — Java path traversal + Rust cmd + Swift ATS #120)
Matcher / corpus follow-ups from PR #100's scan batch
- Recall: open-redirect via Referrer header (Express idiom) #101 + Corpus task: express-referrer-redirect (from scan batch #100) #104 — open-redirect-from-referrer (shipped in matchers: open-redirect-from-referrer + bench fixture #110)
- Recall: nextjs-draft-mode-redirect slug #102 — nextjs-draft-mode-redirect (shipped in matchers: nextjs-draft-mode-redirect + bench fixture #111)
- Matcher: ruby-yaml-unsafe-load (CVE-2013-0156 family) #103 — ruby-yaml-unsafe-load (shipped in matchers: ruby-yaml-unsafe-load (CVE-2013-0156 family) #109)
- Corpus tasks: negative fixtures from flask + jwt-go cores #105 — negative fixtures from flask + jwt-go cores (shipped in corpus: negative fixtures from flask + jwt-go cores #112)
Self-improvement levers
- Prompt evolution loop using the bench corpus + process-compare #86 — Prompt evolution loop using bench corpus (shipped in prompts: --core-prompt evolution loop wiring #122 —
benchsec process-score --core-prompt) - Engine: multi-file taint / control-flow conditions (the dataflow ceiling) #87 — Multi-file taint / control-flow conditions — deferred to v0.3 per
What's not in v0.2
Adoption channels
- Distribution: Homebrew, Scoop, Docker Hub, container registries #88 — Distribution: GoReleaser config publishing Homebrew tap + Scoop bucket + ghcr.io multi-arch images + native packages (shipped in distribution: GoReleaser config for Homebrew + Scoop + ghcr.io #121). Token setup left to repo owner.
Performance follow-up
- perf: cache wazero AST runtime across bench scoreTask calls #114 — Cache wazero AST runtime across bench scoreTask calls — surfaced by corpus: bench tasks for Java + Rust + C (toward #84) #113; works around bumped test timeout. Slips to v0.3.
Exit criteria — final status
- All v0.2 items above ☑ except Engine: multi-file taint / control-flow conditions (the dataflow ceiling) #87 / perf: cache wazero AST runtime across bench scoreTask calls #114, both explicitly deferred to v0.3 with reasoning above.
- 5 multi-repo scan writeups checked in ✓
- ≥ 30 bench tasks ✓ (hit the target exactly)
- An external user has installed via Homebrew/Scoop/Docker — pending tag push + tap repo creation; technically blocked on owner action, not on engineering
- A
SECURITY.mdissue has been opened, triaged, and either fixed or wontfix'd — still pending external reporters
v0.2 ships when
- This issue can be closed once the final PR (prompts: --core-prompt evolution loop wiring #122) merges.
- Two external prerequisites (homebrew tap repo creation; first external install report) are owner-side and don't gate the engineering close.
What's not in v0.2 (deferred to v0.3)
- Multi-file taint engine (Engine: multi-file taint / control-flow conditions (the dataflow ceiling) #87) — substantial RFC-worthy work, slips to v0.3 unless we get free wins via the bench
- Plugin system — separate work, not adoption-blocking
- A web UI — not adoption-blocking; HTML report viewer (HTML report viewer: deepsec report --format html (closes #33) #63 shipped) is enough for v0.2
- Wazero runtime sharing (perf: cache wazero AST runtime across bench scoreTask calls #114) — perf optimization, no user-facing impact
Metadata
Metadata
Assignees
Labels
v0.2Targeted for the v0.2 readiness milestoneTargeted for the v0.2 readiness milestone
Activity
noeljackson commentedon May 14, 2026
v0.2 engineering scope complete.
Shipped this cycle (15 PRs)
Docs / ops: #106 (deepsec doctor), #107 (CONTRIBUTING/SECURITY/index), #108 (deepsec spend).
Matcher / corpus follow-ups from PR #100 scan batch: #109 (ruby-yaml-unsafe-load), #110 (open-redirect-from-referrer + bench fixture), #111 (nextjs-draft-mode-redirect + bench fixture), #112 (negative corpus from flask + jwt-go cores).
Bench corpus growth (#84): #113, #115, #116, #117, #118, #119, #120 — went from 4 → 30 tasks (hit the exit criterion exactly). New language coverage: Java (4 fixtures), Rust (3), C (3), Kotlin (2), Swift (2). Whole-bench recall climbed from baseline 0.92 → 0.97; precision held at 1.0.
Distribution (#88): #121 — GoReleaser config publishes Homebrew tap + Scoop bucket + ghcr.io multi-arch images + .deb/.rpm/.apk packages from a single tag push, per the simpler-than-the-original-issue approach.
Prompt evolution (#86): #122 —
benchsec process-score --core-promptflag plusdocs/prompt-evolution.mdwalkthrough wires the bench corpus into A/B prompt comparison.Deferred to v0.3
Owner-side prerequisites (not gating engineering)
noeljackson/homebrew-deepsecandnoeljackson/scoop-deepsec.HOMEBREW_TAP_TOKEN+SCOOP_TAP_TOKENrepo secrets (fine-grained PAT with Contents R+W on the two tap repos).git tag v0.2.0 && git push origin v0.2.0).Once those are in place,
brew install noeljackson/deepsec/deepsec,scoop install deepsec, anddocker pull ghcr.io/noeljackson/deepsec:v0.2.0all work end-to-end. README install section is already updated to reflect the post-tag reality.🤖 Generated with Claude Code