SigMap is built by a great community of contributors. Thank you to everyone who has helped!
- manojmallick — Core architecture, graph builder, impact analysis, release management
- Claude Code — AI-assisted development, code generation, testing
- ContextForge — Integration, adapters, multi-framework support
- David Schoch — GDScript extractor, language support
- Sean Campbell — Framework detection, configuration
- Denis Solonenko — GDScript extractor (#146)
- Matt Van Horn — Testing, reliability improvements
- kumamaki — Bug fixes, improvements
- SigMap Bot — Release automation, version management
- Community members — Bug reports, feedback, discussions
When you contribute to SigMap, your GitHub account will appear in the contributors graph if:
- Your commit author email is linked to your GitHub account profile
- You make commits using that email address
To ensure proper attribution:
- Link your email to your GitHub account: https://github.com/settings/emails
- Configure git locally:
git config user.email your-github-email@example.com
We welcome contributions! See Contributing for guidelines.
- @manojmallick — feat(extractors): line anchors for Kotlin/Swift/PHP/Scala/Dart — 9 brace languages now anchored (#486, PR #487)
- @manojmallick — feat(retrieval): route surface-enrichment, opt-in
retrieval.surfaceEnrichment, measured +0 on the A/B so default stays off; newbenchmark:surface-enrichmentgate (#488, PR #489) - @manojmallick — docs: "Your agent's live loop" framing in the MCP guide + README (#490, PR #491)
- @manojmallick — feat(extractors): line anchors for Java, Go, Rust, and C# — newline-preserving comment strips + brace-matched
:start-endranges; Evidence Pack anchorCoverage 0→1.0 on those languages (#483, PR #484)
- @manojmallick — fix(bench): hermetic grounding benchmark — snapshot/restore of repo context artifacts stops cross-suite pollution (A/B 32.2% → true 87.8%); 3 regression tests (#480, PR #481)
- @manojmallick — feat(evidence): Evidence Pack schema v2 — published JSON Schema, multi-factor risk labels, measured test-discovery provenance (F1 0.98, guard-tested), generator identity (#477, PR #478)
- @manojmallick — feat(retrieval): call-graph ranking boost — opt-in
retrieval.callGraphBoost, measured +0 on the 90-task A/B so the default stays off; newbenchmark:callgraph-boostgate script (#474, PR #475)
- @manojmallick — feat(graph): call-graph support for Java, Go, and Rust (GR1) — per-language def extractors, lifetime-safe Rust masker, Go/Java same-package resolution scope; all call-graph consumers inherit the languages (#471, PR #472)
- @manojmallick — feat(graph): method-level blast-radius scoring (GR2) — deterministic per-file score/tier from the call graph, wired into
review-prfindings, the PR Evidence report, and the newget_method_impactMCP tool (19→20 tools) (#468, PR #469)
- @manojmallick — feat(wiki):
sigmap wiki— deterministic architecture narrative from signatures + dependency graph (no LLM, byte-stable,--json/--out), closing D9, the last unstarted in-boundary master-plan item (#465, PR #466)
- @manojmallick — feat(format): terse signature encoder — opt-in
--tersedeterministic compaction of the signature block, measured −16.1% on this repo via the newbenchmark:tersemeasure-first gate; line anchors preserved byte-exactly, ranker parse-back regression-tested (#462, PR #463) - @manojmallick — docs(readme): reliable Star History link (PR #461); 'Verified on MseeP' badge (PR #460)
- @manojmallick — feat(cli): honesty fixes across the CLI surface — claim-level grounding in
judge, NL retrieval-confidence invalidate --query, realplanimpact (+ depth-cap bug fix), content-based secret scan inreview-pr, unifiedgainpricing, auditable--healthcomposite, tsconfig-alias import-graph resolution, single-wordconventionsfix, diff-drivensuggest-profile, and visible extractor truncation (#457, PR #458) - @mseep-ai — docs: add the MseeP.ai security assessment badge to the README (PR #456)
- @manojmallick — ci(pages): retry the GitHub Pages deploy once on GitHub's transient "try again later", so releases stop going red on a backend timing flake (CI only; package unchanged) (#451)
- @manojmallick — feat(daemon): detached watch daemon —
sigmap daemon start|stop|statusruns--watchas a managed background process (PID + log under.context/), zero-dependency and shell-free (D1) (#447, PR #448)
- @manojmallick — fix: close the
gen-contextdeterminism residual — replaced theDate.now()recency boost with a deterministic monotonic counter, so output is byte-stable across all 43 benchmark repos and retrieval hit@5 reproduces at 87.8%; adds a byte-equality regression guard (#440, PR #444) - @manojmallick — fix(meta): count the relocated Python test in its new
test/location; docs/CI: serve OG banner + Search Console file, run Python tests in CI; chore: repo hygiene (drop orphaned GIFs, relocate stray test)
- @manojmallick — feat(mcp): expose the squeeze engine —
squeeze_outputMCP tool (18 → 19 tools) for mid-session output compression +sigmap squeeze --response <file|->CLI flag (D6) (#437, PR #438) - @manojmallick — fix: harden
gen-contextoutput determinism — filePath tie-break in the token-budget drop-order, sorted source walk, and tie-broken source-root detection so the generated context is byte-stable (#440, PR #441)
- @manojmallick — docs(benchmark): multi-model cost savings — quality benchmark now reports GPT-4o + Claude Sonnet + Claude Haiku input-cost savings; corrected stale Claude pricing (Haiku $0.80→$1.00, Opus $15→$5) to verified 2026-07 rates (#433, PR #434)
- @manojmallick — feat(graph): method/caller-level call-graph (D4 v1) — symbol-level edges (
buildCallGraph) + method blast radius (methodImpact) for JS/TS + Python, with--callers/--calleesCLI (#429, PR #430)
- @manojmallick — feat(grounding): Phase 1 "bank the A" — public reproducible benchmark harness (
public-benchmarks/), installed dependency version pins in the context header (D8), andsigmap verifypromoted to the documented grounding flagship (#425, PR #426)
- @manojmallick — feat(retrieval): deterministic query expansion — curated synonym/abbreviation bridge for the BM25 ranker (auth↔authentication, db↔database, …) at a discount weight; a vocabulary-mismatch recall aid, benchmark-neutral (no hit@5 regression) (#421, PR #422)
- @manojmallick — feat(review): PR Evidence Report (v9.0 G3) — branded, deterministic Markdown review artifact (signatures + blast radius + related tests + risk labels + review-pr findings);
sigmap review-pr --markdown, CI-gateable, no LLM (#417, PR #418)
- @manojmallick — feat(verify): Python site-packages grounding — extend the local-library moat (G5/D5) to venv libraries; verify AI-suggested Python code against installed packages'
__init__.py/.pyiexports + versions (D8), no Python runtime (#413, PR #414)
- @manojmallick — feat(mcp):
verify_suggestiontool (18th) — expose the local-library grounding moat to agents; verify an AI code suggestion against repo + installed-library symbols before writing, with pinned versions (D8) (#409, PR #410)
- @manojmallick — feat(verify): local-library signature index (v9.0 G5/D5) — verify AI suggestions against the libraries actually installed in
node_modules(direct-dep.d.tsexports + D8 version pinning), the private-API grounding moat; genuine library calls stop false-flagging as fake-symbol (#405, PR #406)
- @manojmallick — feat(v8.5): repo-context coverage — env-schema / build-CI / config-manifest / DB-migration map analyzers (C1); measured cross-language test discovery + reproducible benchmark, F1 98.0% / hit@1 97.4% (C2); richer precedence-ordered risk labels — migration/payment/auth/public-api (C3) (#401, PR #402)
- @manojmallick — docs: correct the answer-correctness multiplier badge in the comparison chart (×5.2 → ×6.8) (PR #399)
- @manojmallick — feat(positioning): reposition every public surface to "the deterministic, verifiable grounding layer for AI code work" + agent recipes (v8.0 E2/E4); documents the shipped
evidence/doctorcommands; adds a repositioning gate (#389, PR #390)
- @manojmallick — feat(mcp): one-command per-client MCP install —
sigmap mcp install <client>+sigmap mcp list(v8.0 E4); creates the config when absent, idempotent, handles mcpServers / Zed / Codex shapes;--globalscope (#385, PR #386)
- @manojmallick — feat(doctor): add
sigmap doctor(v8.0 E3) — one-shot setup diagnostic (config · context · index · freshness · coverage · MCP wiring) with actionable fixes;--json; CI-usable exit codes (#381, PR #382) - @manojmallick — ci(sync): self-heal
developwhen the release PR merge auto-deletes it (#380)
- @manojmallick — feat(mcp): add
get_diff_context&get_architecture_overviewMCP tools (v8.0 D3) — changed-file signatures + blast radius, and a one-call architecture map; MCP surface 15→17 (#376, PR #377)
- @manojmallick — feat(evidence): Evidence Pack JSON v1 — deterministic, machine-consumable signature+evidence map (
sigmap evidence) with JSON + Markdown handoff modes, byte-stable output, and a sha256 grounding hash (#372, PR #373)
- @manojmallick — docs: surface the StarMapper stargazer map (517 stars · 37 countries) — README badge + Support link, docs community link, README-structure test (#360, PR #361)
- @manojmallick — docs: publish the first measured §9 grounding result (
version.jsonablation) — 5×100 repo-fact tasks on Gemini: grounding cut flagged codebase-fact errors from 99.8 [99–100] to 0.2 [0–1] per 100 (factual-recall grounding)
- @manojmallick — feat: redesign the §9 ablation corpus as checkable repo-fact questions (which file defines
<name>, what params) — a wrong path is a checkable hallucination and example code is forbidden, so the metric isolates grounding instead of guard precision; idscall-→fact-(#356, PR #357)
- @manojmallick — feat: robust §9 ablation —
--runs Naveraging (mean ± range) via pureaggregateRuns, 100-task real-symbol corpus (was 40); makes one invocation yield a statistically stable grounding number (#353, PR #354)
- @manojmallick — fix: clear the last two
verify-ai-outputfalse-positive classes — camelCase placeholders (myExample.js) and documentation-placeholder imports (@scope/utils,some-module,./local-file); ordinary words and genuine fakes still flag; exposes the true §9 grounding delta (#350, PR #351)
- @manojmallick — fix: harden
verify-ai-outputfile-path extractor — skip runtime/library product names (Node.js,Next.js, …) and illustrative placeholders (example.js,minimal-example.js); genuine repo paths still flag; removes the dominant Hallucination Guard false-positive class (#347, PR #348)
- @manojmallick — feat: realistic §9 ablation — real-symbol corpus (
gen-ablation-corpus.mjs), exact-signature grounding,--verboseflagged items;scoreAnswerDetail(#344, PR #345); fix: default Gemini model → gemini-2.5-flash (#343)
- @manojmallick — feat: Gemini (AI Studio) provider for the §9 LLM A/B ablation runner — auto-detected from GEMINI_API_KEY;
--provider/--modelflags (#340, PR #341)
- @manojmallick — feat:
sigmap --initwrites a Creation workflow block into CLAUDE.md (renderCreationWorkflowBlock/injectCreationWorkflow); final IMPL item — grounded-codegen plan complete (#337, PR #338)
- @manojmallick — feat: scaffold persistence — write an accepted proposal to
.context/scaffold/latest.md;renderScaffoldMarkdown/scaffoldPath; Gap 2 §6.2 (#334, PR #335)
- @manojmallick — feat:
sigmap conventions --update— incremental rescan (refresh snapshot only when source changed);changedSince/planUpdate; completes the §4 flag set (#331, PR #332)
- @manojmallick — feat:
sigmap conventions --fix— exhaustive rename/move checklist (every offending file, full from→to paths);buildFixList; Layer 3 (#328, PR #329)
- @manojmallick — feat: LLM A/B hallucination ablation harness (IMPL §9) — injected-completer harness (
buildGrounding/scoreAnswer/runAblation) + live runner; offline-testable, network confined to scripts/ (#325, PR #326)
- @manojmallick — feat:
sigmap conventions --ci— CI gate on overall convention consistency (--min,--no-regress);ciGate; Layer 3 (#322, PR #323)
- @manojmallick — feat:
sigmap conventions --report— consistency audit + overall score + trend vs last run;scoreReport/snapshot; Layer 3 (#319, PR #320)
- @manojmallick — feat:
sigmap create— orchestrate the 4-stage grounded-creation pipeline (scaffold → verify-plan → verify-ai-output → review-pr) with n/4 numbering;orchestrate; Gap 2 capstone (#316, PR #317)
- @manojmallick — feat:
sigmap review-pr— diff audit (scope drift, god-node edits, missing tests, security-sensitive files);reviewPr; last guard stage of the create pipeline (Gap 2) (#313, PR #314)
- @manojmallick — feat:
sigmap verify-plan— check a plan against the live index (files/symbols exist, blast radius, scope) before execution;verifyPlan; Gap 2 grounded codegen (#310, PR #311)
- @manojmallick — feat:
sigmap scaffold— convention-matched proposal with a confidence floor (soft threshold 0.70, non-overridable hard floor 0.50);proposeScaffold; Layer 4 grounded codegen (#307, PR #308)
- @manojmallick — feat:
sigmap conventions --inject— write the auto-detected conventions block into CLAUDE.md (idempotent, marker-scoped);renderConventionsBlock/injectConventions; Layer 3 grounded codegen (#304, PR #305)
- @manojmallick — feat:
sigmap conventions --conflicts— per-convention breakdown (counts, bars, example files) + rename suggestions;analyzeConflicts/toNamingStyle; example-file tracking inscoreConvention; Layer 3 grounded codegen (#301, PR #302)
- @manojmallick — feat:
sigmap conventions— extract & report a repo's coding conventions (file naming, export style, test framework) for TS/JS/Python; reusablescoreConventionconsistency scorer; Layer 3 grounded codegen (#298, PR #299)
- @manojmallick — fix: shell-free subprocess calls (zero
execSyncin the published surface),main→ core API, removed unused device fingerprint, star nudge counts plainsigmapruns (#250, PRs #251 #252)
- @manojmallick — feat:
verify-ai-outputHallucination Guard prototype — deterministic fake-file / fake-import / fake-symbol detectors, markdown +--json, offline (#227, PR #228)
- @manojmallick — feat: line anchors for JavaScript + member-level anchors (TS & JS); index-mode token cut 4.6% → 32–42% on real repos; overhead-aware token budget (#223, PR #224)
- @manojmallick — feat: Surgical Context Phase 2 —
get_linesMCP tool,ask --mode index,ask --since, budget-aware body collapse, Token Reduction dashboard panel (#219, PR #220) - @manojmallick — ci: add
workflow_dispatchto the develop→main sync workflow (PR #218)
- @rudi193-cmd — fix: include hot-cold cold signatures in the bundled MCP server (#201, PR #216)
- Line anchors (
:start-end) on TypeScript & Python signatures — Surgical Context Phase 1 - MCP registration with intelligent fallback to
.claude/settings.json
- MCP cache improvements for hot/cold context indexing
- Binary build reliability with r-manifest module bundling
- npm publish workflow with automation token support
- Test assertion fixes and integration test reliability
- Documentation updates for 31 supported languages (R + GDScript)
- Cross-platform compatibility (macOS, Linux, Windows)
Thank you to all contributors! Your work makes SigMap better for everyone. 🙏