sigmap 6.15.0 → 7.0.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -10,6 +10,42 @@ Format: [Semantic Versioning](https://semver.org/)
10
10
 
11
11
  ---
12
12
 
13
+ ## [7.0.1] — 2026-06-14
14
+
15
+ Patch release — supply-chain hardening and package hygiene, plus a wider star nudge.
16
+
17
+ ### Fixed
18
+ - **Eliminated system-shell access (#252):** every `child_process.execSync` call (which runs via `/bin/sh -c`) was converted to shell-free `execFileSync` with an arguments array. Several commands had previously interpolated values into the command string (`git diff ${range}`, `HEAD~${n}`, `printf '%s' … | ${clipCmd}`, `node -e "…http.get…"`) — a real shell-injection surface. New `src/util/git.js` (`git()`/`tryGit()`) centralizes shell-free git; the `extends` config fetch passes the URL as an argv to node; `compare` spawns node by argv; clipboard copy writes via stdin. Net: zero `execSync`/`exec`/`shell:true` in the published surface, which clears Socket's "Shell access" capability alert.
19
+
20
+ ### Changed
21
+ - **Star nudge now counts plain `sigmap` runs (#251):** the one-time GitHub-star nudge previously only counted `ask`/`squeeze`. Users who only run `sigmap` to generate the context file now also reach the 10-run threshold. Counted once per process at the end of generation (monorepo-safe, tracked at the repo root); interactive-only — silent under `--json`/`--report`/`--quiet`/non-TTY.
22
+ - **`main` points at the importable core (#252):** `package.json` `main` changed from the CLI bundle (`gen-context.js`, which runs `main()` + exits on `require`) to `packages/core/index.js`, matching `exports["."]`. Bundlephobia and legacy resolvers now see the real zero-dep API.
23
+ - **Removed unused device fingerprint (#252):** the star nudge no longer records `machineId = sha256(os.hostname())` in `.context/usage.json` — it was never read or transmitted. Dropped the now-unused `os`/`crypto` requires.
24
+
25
+ ---
26
+
27
+ ## [7.0.0] — 2026-06-14
28
+
29
+ Major release — **Squeeze** makes `ask` minimize pasted input by default, a behavioral change to the core command.
30
+
31
+ ### Added
32
+ - **Squeeze — input minimization (#238):** `sigmap ask` now classifies a pasted blob (stack trace / CI log / JSON) and minimizes it before ranking — deduping frames, stripping vendor noise, collapsing repeated array items — and **enriches the top stack frame** with its real signature from the symbol index. New `sigmap squeeze <file|->` command and `--squeeze` (auto-accept), `--no-squeeze`, `--squeeze-threshold N` flags. New `src/squeeze/{classify,cilog,stacktrace,jsonpayload,index}.js`; zero-dep, deterministic, offline.
33
+ - **Star Nudge (#238):** a one-time, race-safe GitHub-star prompt after ≥10 runs / ≥8 successes (`.context/usage.json`).
34
+ - **`llms.txt` + `llms-full.txt` generator (#243):** SigMap's own LLM reference is generated from source of truth (MCP tools, config keys, languages, `version.json` metrics, CLI help), validated in CI so it can never go stale, and published to the docs site + repo root. `npm run generate:llms` / `validate:llms`; `prepublishOnly` regenerates before publish.
35
+ - **npm discoverability + GitHub Sponsors (#241):** benefit-driven description, 20 keywords, `funding` field + `.github/FUNDING.yml`.
36
+
37
+ ### Changed
38
+ - **BREAKING: `sigmap ask` now classifies its input and may prompt** to minimize large pasted stack traces / logs / JSON before ranking. Interactive only — piped/CI usage is unaffected (no prompt), and `--no-squeeze` fully disables it.
39
+ - **Token budget keeps full signatures (#240):** when context exceeds `maxTokens`, low-priority files are dropped (and only marginal overflow collapses to anchors) — signatures keep their parameters/return types instead of being gutted to bare line anchors. The repo's own context config raises `maxTokens` (auto-scaling off) so it fits all files with full signatures.
40
+ - **One consistent usage block (#240):** every generated context file (CLAUDE.md / AGENTS.md / copilot-instructions.md / …) now carries a single canonical `## SigMap commands` block emitted from `formatOutput`; the redundant `## Tools` JSON in AGENTS.md was removed.
41
+ - **Benchmark repos pinned to fixed commits (#236):** retrieval/token benchmarks now clone pinned SHAs, so metrics move only when SigMap changes — not when upstream repos do.
42
+
43
+ ### Fixed
44
+ - **Signatures no longer gutted under budget (#240):** the headline regression in the generated context files is resolved (see Changed).
45
+ - **prdiff symbol-name extraction (#247):** the changes block no longer emits phantom 2-char fragments (e.g. `+is`/`~is`); `extractName` now handles `export class`, `const x = () =>`, async/visibility modifiers, and returns nothing for re-export lines.
46
+
47
+ ---
48
+
13
49
  ## [6.15.0] — 2026-06-09
14
50
 
15
51
  ### Added
package/README.md CHANGED
@@ -48,9 +48,9 @@ SigMap extracts function and class signatures from your codebase and feeds the r
48
48
  ## Why SigMap?
49
49
 
50
50
  - **75.6% hit@5** — right file found in top 5 results (vs 13.6% baseline)
51
- - **97.1% token reduction** — average across 21 real repos
52
- - **51.1% task success rate** — up from 10% without context
53
- - **1.73 prompts per task** — down from 2.84 (39.0% fewer retries)
51
+ - **97.0% token reduction** — average across 21 real repos
52
+ - **52.2% task success rate** — up from 10% without context
53
+ - **1.72 prompts per task** — down from 2.84 (39.4% fewer retries)
54
54
  - **31 languages supported** — TypeScript, Python, Go, Rust, Java, R, and 25 others
55
55
  - **No vendor lock-in** — works with any AI assistant or local LLM
56
56
  - **No API costs** — use local models (Ollama, llama.cpp, vLLM) with zero token fees
@@ -87,13 +87,13 @@ Ask → Rank → Context → Validate → Judge → Learn
87
87
  ## Benchmark
88
88
 
89
89
  ```
90
- Benchmark : sigmap-v6.15-main (21 repositories, including R language)
91
- Date : 2026-06-09
90
+ Benchmark : sigmap-v7.0-main (21 repositories, including R language)
91
+ Date : 2026-06-14
92
92
 
93
93
  Hit@5 : 75.6% (baseline 13.6% — 5.6× lift)
94
- Token reduction: 97.1% (across 21 repos)
95
- Prompt reduction : 39.0% (2.84 → 1.73 prompts per task)
96
- Task success : 51.1% (baseline 10%)
94
+ Token reduction: 97.0% (across 21 repos)
95
+ Prompt reduction : 39.4% (2.84 → 1.72 prompts per task)
96
+ Task success : 52.2% (baseline 10%)
97
97
  Repos tested : 21 (JavaScript, Python, Go, Rust, Java, R, C++, C#, Dart, Swift, Ruby, PHP, Scala, Kotlin, and more)
98
98
  ```
99
99