@sema-agent/server 7.75.1 → 7.75.2

package/dist/trace/redact.d.ts

@@ -30,6 +30,8 @@
  *
  * 两组臂,顺序即优先序:{@link FULL_SECRET_PATTERNS}(完整形)在前,{@link FRAGMENT_SECRET_PATTERNS}
  * ([ref] 前缀残片,上游先截后脱的产物 ⇒ 带 `:partial` 后缀的 token)在后。
+ * ⚠️ **一个例外,自 7.75.2**:完整臂里的**裸形段**内部没有优先序 —— 该段全臂一次收候选跨度、解重叠、一次替换
+ * ({@link shapeSectionRule}),段内臂序对输出无语义。段**之间**的先后照旧是契约。
  */
 /** 本模块替换 token 的**闭形**正则源(`«redacted»` / `«redacted:aws-key»` / `«redacted:gh-token:partial»` …):
  *  URL-userinfo 臂用它把「用户名位上已经是本模块 token」整个当一个用户名单元吃掉。模块加载时对

package/dist/trace/redact.js

@@ -207,32 +207,65 @@ function forgetEscapedSeparatorTails() {
 }
 const VALUE_ARM_STICKY = new Map();
 const VALUE_ARM_BODIES = [];
-const valueArm = (prefixSrc, valueClass, min) => {
+const valueArm = (prefixSrc, valueClass, min, evidence) => {
     const body = `${prefixSrc}${twoStrongAhead(valueClass)}${atLeast(valueClass, min)}`;
-    const g = new RegExp(`(?<!${WORD_INTERIOR_CLASS})${body}`, "g");
-    VALUE_ARM_STICKY.set(g, new RegExp(body, "y"));
     VALUE_ARM_BODIES.push(body);
-    return g;
+    switch (evidence) {
+        case "self-evident":
+            return new RegExp(body, "g");
+        case "needs-left-boundary": {
+            const g = new RegExp(`(?<!${WORD_INTERIOR_CLASS})${body}`, "g");
+            VALUE_ARM_STICKY.set(g, new RegExp(body, "y"));
+            return g;
+        }
+        default: {
+            const unhandled = evidence;
+            throw new Error(`redact: unhandled value-arm evidence ${String(unhandled)}`);
+        }
+    }
 };
 let anyValueArm;
 function anyValueArmRe() {
     return (anyValueArm ??= new RegExp(`(?:${VALUE_ARM_BODIES.join("|")})`, "y"));
 }
-function runValueArm(g, y, repl, text, wantCount) {
+function collectRegexSpans(g, repl, order, text, out) {
+    g.lastIndex = 0;
+    for (let m = g.exec(text); m !== null; m = g.exec(text)) {
+        const end = m.index + m[0].length;
+        out.push({ start: m.index, end, repl, order });
+        g.lastIndex = end > m.index ? end : m.index + 1;
+    }
+}
+function resolveSpanOverlaps(spans) {
+    spans.sort((a, b) => a.start - b.start || b.end - a.end || a.order - b.order);
+    const kept = [];
+    let covered = 0;
+    for (const s of spans) {
+        if (s.end <= covered)
+            continue;
+        kept.push(s.start >= covered ? s : { ...s, start: covered });
+        covered = s.end;
+    }
+    return kept;
+}
+function applySpans(text, spans) {
+    if (spans.length === 0)
+        return { out: text, count: 0 };
+    const parts = [];
+    let copied = 0;
+    for (const s of spans) {
+        parts.push(text.slice(copied, s.start), s.repl);
+        copied = s.end;
+    }
+    parts.push(text.slice(copied));
+    return { out: parts.join(""), count: spans.length };
+}
+function collectValueArmSpans(g, y, repl, order, text, out) {
     const tails = escapedSeparatorTails(text);
     if (tails === undefined) {
-        if (!wantCount)
-            return { out: text.replace(g, repl), count: 0 };
-        let hits = 0;
-        const replaced = text.replace(g, () => {
-            hits += 1;
-            return repl;
-        });
-        return { out: replaced, count: hits };
+        collectRegexSpans(g, repl, order, text, out);
+        return;
     }
-    let out;
-    let copied = 0;
-    let count = 0;
     let pos = 0;
     let ti = 0;
     let bIdx = -1;
@@ -275,15 +308,14 @@ function runValueArm(g, y, repl, text, wantCount) {
         }
         else
             break;
-        (out ??= []).push(text.slice(copied, idx), repl);
-        copied = end;
-        count += 1;
+        out.push({ start: idx, end, repl, order });
         pos = end;
     }
-    if (out === undefined)
-        return { out: text, count: 0 };
-    out.push(text.slice(copied));
-    return { out: out.join(""), count };
+}
+function runValueArm(g, y, repl, text) {
+    const spans = [];
+    collectValueArmSpans(g, y, repl, 0, text, spans);
+    return applySpans(text, spans);
 }
 const WORD_SEGMENT_CLASS = "[A-Za-z0-9._~+/-]";
 const keywordArm = (keywordSrc, valueClass, min, tailSrc = "") => new RegExp(`(?:${keywordSrc}\\s+${atLeast(valueClass, min)}${tailSrc}` +
@@ -303,7 +335,6 @@ const KEYWORD_SECRET_PATTERNS = [
     [keywordArm("[Tt]oken", "[A-Za-z0-9._~+/-]", 20), "token «redacted»"],
 ];
 const SHAPE_SECRET_PATTERNS = [
-    [valueArm("sk-", "[A-Za-z0-9_-]", 16), "«redacted:key»"],
     [new RegExp(String.raw `(?:ghp|gho|ghu|ghs|ghr)_${atLeast("[A-Za-z0-9]", 20)}`, "g"), "«redacted:gh-token»"],
     [new RegExp(String.raw `github_pat_[A-Za-z0-9]{22}_${atLeast("[A-Za-z0-9]", 59)}`, "g"), "«redacted:gh-token»"],
     [new RegExp(String.raw `xox[baprs]-${atLeast("[A-Za-z0-9-]", 10)}`, "g"), "«redacted:slack-token»"],
@@ -311,6 +342,7 @@ const SHAPE_SECRET_PATTERNS = [
     [new RegExp(String.raw `ASIA${atLeast("[0-9A-Z]", 16)}`, "g"), "«redacted:aws-key»"],
     [new RegExp(String.raw `(?:gl(?:pat|dt|rt|rtr|cbt|ptt|oas|ft|imt|agent|wt|soat|ffct)-${atLeast("[A-Za-z0-9_-]", 20)}|(?:sk|rk)_(?:live|test|org)_${atLeast("[A-Za-z0-9]", 24)}|pypi-${atLeast("[A-Za-z0-9_-]", 85)})`, "g"), "«redacted:key»"],
     [new RegExp(String.raw `xapp-\d-${atLeast("[A-Za-z0-9-]", 20)}`, "g"), "«redacted:slack-token»"],
+    [valueArm("sk-", "[A-Za-z0-9_-]", 16, "self-evident"), "«redacted:key»"],
 ];
 const FULL_SECRET_PATTERNS = [...STRUCTURAL_SECRET_PATTERNS, ...KEYWORD_SECRET_PATTERNS, ...SHAPE_SECRET_PATTERNS];
 const PEM_HEADER_IDENT_MAX = 31;
@@ -688,14 +720,14 @@ function redactHeadlessPpkBodies(text) {
     return count === 0 ? { out: text, count: 0 } : { out: out + text.slice(last), count };
 }
 const FRAGMENT_SECRET_PATTERNS = [
-    [valueArm("sk-", "[A-Za-z0-9_-]", 6), "«redacted:key:partial»"],
-    [valueArm("(?:ghp|gho|ghu|ghs|ghr)_", "[A-Za-z0-9]", 8), "«redacted:gh-token:partial»"],
+    [valueArm("sk-", "[A-Za-z0-9_-]", 6, "needs-left-boundary"), "«redacted:key:partial»"],
+    [valueArm("(?:ghp|gho|ghu|ghs|ghr)_", "[A-Za-z0-9]", 8, "needs-left-boundary"), "«redacted:gh-token:partial»"],
     [new RegExp(String.raw `github_pat_${atLeast("[A-Za-z0-9_]", 8)}`, "g"), "«redacted:gh-token:partial»"],
-    [valueArm("xox[baprs]-", "[A-Za-z0-9-]", 4), "«redacted:slack-token:partial»"],
-    [valueArm("AKIA", "[0-9A-Z]", 6), "«redacted:aws-key:partial»"],
-    [valueArm("ASIA", "[0-9A-Z]", 8), "«redacted:aws-key:partial»"],
-    [valueArm("(?:gl(?:pat|dt|rt|rtr|cbt|ptt|oas|ft|imt|agent|wt|soat|ffct)-|(?:sk|rk)_(?:live|test|org)_|pypi-)", "[A-Za-z0-9_-]", 8), "«redacted:key:partial»"],
-    [valueArm("xapp-\\d-", "[A-Za-z0-9-]", 4), "«redacted:slack-token:partial»"],
+    [valueArm("xox[baprs]-", "[A-Za-z0-9-]", 4, "needs-left-boundary"), "«redacted:slack-token:partial»"],
+    [valueArm("AKIA", "[0-9A-Z]", 6, "needs-left-boundary"), "«redacted:aws-key:partial»"],
+    [valueArm("ASIA", "[0-9A-Z]", 8, "needs-left-boundary"), "«redacted:aws-key:partial»"],
+    [valueArm("(?:gl(?:pat|dt|rt|rtr|cbt|ptt|oas|ft|imt|agent|wt|soat|ffct)-|(?:sk|rk)_(?:live|test|org)_|pypi-)", "[A-Za-z0-9_-]", 8, "needs-left-boundary"), "«redacted:key:partial»"],
+    [valueArm("xapp-\\d-", "[A-Za-z0-9-]", 4, "needs-left-boundary"), "«redacted:slack-token:partial»"],
 ];
 const REDACTION_TOKENS = new Set([...FULL_SECRET_PATTERNS.map(([, r]) => r), PEM_FRAGMENT_TOKEN, ...FRAGMENT_SECRET_PATTERNS.map(([, r]) => r)]
     .flatMap((r) => r.match(/«[^»]*»/g) ?? []));
@@ -715,6 +747,8 @@ function redactionTokenEndAt(s, p, limit) {
     return 0;
 }
 const labelOfToken = (token) => token.replace(/[«»]/g, "").replace(/^redacted:/, "");
+const NO_HITS = [];
+const singleHit = (label, arm, r) => r.count > 0 ? { out: r.out, hits: [{ label, count: r.count, arm }] } : { out: r.out, hits: NO_HITS };
 const regexRule = ([re, repl], arm) => {
     const groupCount = (new RegExp(`${re.source}|`).exec("")?.length ?? 1) - 1;
     const refs = [...repl.matchAll(/\$(.)/g)].map(([, d]) => d ?? "");
@@ -726,23 +760,20 @@ const regexRule = ([re, repl], arm) => {
     const sticky = VALUE_ARM_STICKY.get(re);
     if (sticky !== undefined && refs.length > 0)
         throw new Error(`redact: value arm replacement ${JSON.stringify(repl)} must not use $N refs`);
-    return {
-        label: labelOfToken(repl),
-        arm,
-        run: (text, wantCount, json) => {
-            if (sticky !== undefined)
-                return runValueArm(re, sticky, repl, text, wantCount);
-            const rx = re;
-            if (!wantCount)
-                return { out: text.replace(rx, repl), count: 0 };
-            let n = 0;
-            const out = text.replace(rx, (...args) => {
-                n += 1;
-                return expand === undefined ? repl : expand(args);
-            });
-            return { out, count: n };
-        },
+    const label = labelOfToken(repl);
+    const applyOnce = (text, wantCount) => {
+        if (sticky !== undefined)
+            return runValueArm(re, sticky, repl, text);
+        if (!wantCount)
+            return { out: text.replace(re, repl), count: 0 };
+        let n = 0;
+        const out = text.replace(re, (...args) => {
+            n += 1;
+            return expand === undefined ? repl : expand(args);
+        });
+        return { out, count: n };
     };
+    return { run: (text, wantCount) => singleHit(label, arm, applyOnce(text, wantCount)) };
 };
 const URL_USERINFO_PATTERN = (() => {
     const row = FULL_SECRET_PATTERNS.find(([, repl]) => repl.endsWith("»@"));
@@ -937,16 +968,43 @@ function redactPpkContainers(text, record) {
     }
     return count === 0 ? { out: text, count: 0 } : { out: out + text.slice(last), count };
 }
-const urlUserinfoRule = { label: labelOfToken(URL_USERINFO_PATTERN[1]), arm: "full", run: (text, _wantCount, json) => redactUrlUserinfo(text, json) };
+const URL_USERINFO_LABEL = labelOfToken(URL_USERINFO_PATTERN[1]);
+const urlUserinfoRule = { run: (text, _wantCount, json) => singleHit(URL_USERINFO_LABEL, "full", redactUrlUserinfo(text, json)) };
+const shapeSectionRule = (() => {
+    const arms = SHAPE_SECRET_PATTERNS.map(([re, repl], order) => {
+        if (/\$./.test(repl))
+            throw new Error(`redact: shape-section replacement ${JSON.stringify(repl)} must not use $N refs`);
+        return { g: re, y: VALUE_ARM_STICKY.get(re), repl, order };
+    });
+    return {
+        run: (text, wantCount) => {
+            const spans = [];
+            for (const a of arms) {
+                if (a.y === undefined)
+                    collectRegexSpans(a.g, a.repl, a.order, text, spans);
+                else
+                    collectValueArmSpans(a.g, a.y, a.repl, a.order, text, spans);
+            }
+            const kept = resolveSpanOverlaps(spans);
+            const out = applySpans(text, kept).out;
+            if (!wantCount || kept.length === 0)
+                return { out, hits: NO_HITS };
+            const byToken = new Map();
+            for (const sp of kept)
+                byToken.set(sp.repl, (byToken.get(sp.repl) ?? 0) + 1);
+            return { out, hits: [...byToken].map(([token, count]) => ({ label: labelOfToken(token), count, arm: "full" })) };
+        },
+    };
+})();
 const SECRET_RULES = [
     ...STRUCTURAL_SECRET_PATTERNS.map((p) => (p === URL_USERINFO_PATTERN ? urlUserinfoRule : regexRule(p, "full"))),
-    { label: labelOfToken(PPK_CONTAINER_TOKEN), arm: "full", run: (text, _wantCount, _json, observe) => redactPpkContainers(text, observe) },
-    { label: labelOfToken(PEM_FRAGMENT_TOKEN), arm: "fragment", run: (text, _wantCount, json) => redactPemFragments(text, json) },
-    { label: labelOfToken(HEADLESS_PPK_TOKEN), arm: "fragment", run: (text) => redactHeadlessPpkBodies(text) },
+    { run: (text, _wantCount, _json, observe) => singleHit(labelOfToken(PPK_CONTAINER_TOKEN), "full", redactPpkContainers(text, observe)) },
+    { run: (text, _wantCount, json) => singleHit(labelOfToken(PEM_FRAGMENT_TOKEN), "fragment", redactPemFragments(text, json)) },
+    { run: (text) => singleHit(labelOfToken(HEADLESS_PPK_TOKEN), "fragment", redactHeadlessPpkBodies(text)) },
     ...KEYWORD_SECRET_PATTERNS.map((p) => regexRule(p, "full")),
-    ...SHAPE_SECRET_PATTERNS.map((p) => regexRule(p, "full")),
+    shapeSectionRule,
     ...FRAGMENT_SECRET_PATTERNS.map((p) => regexRule(p, "fragment")),
-    { ...urlUserinfoRule, run: (text, _wantCount, json) => ({ out: redactUrlUserinfo(text, json).out, count: 0 }) },
+    { run: (text, _wantCount, json) => ({ out: redactUrlUserinfo(text, json).out, hits: NO_HITS }) },
 ];
 let onRedaction;
 export function setRedactionObserver(fn) {
@@ -958,8 +1016,8 @@ function applyRules(text, wantCount, json, observe) {
     for (const rule of SECRET_RULES) {
         const r = rule.run(out, wantCount, json, observe);
         out = r.out;
-        if (wantCount && r.count > 0)
-            hits.push({ label: rule.label, count: r.count, arm: rule.arm });
+        if (wantCount && r.hits.length > 0)
+            hits.push(...r.hits);
     }
     return { out, hits };
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sema-agent/server",
-  "version": "7.75.1",
+  "version": "7.75.2",
   "description": "Sema Server — the server/API implementation layer for Sema, wiring core, registry, model providers, and cloud agent execution. Built on @sema-agent/core.",
   "type": "module",
   "license": "BUSL-1.1",
@@ -54,7 +54,7 @@
     "build:binary:run-local:darwin-arm64": "bun build --compile --target=bun-darwin-arm64 src/run-local.ts --outfile dist/run-local-darwin-arm64"
   },
   "dependencies": {
-    "@sema-agent/core": "7.17.4",
+    "@sema-agent/core": "7.17.5",
     "@sema-agent/settings-schema": "1.10.0",
     "e2b": "^2.28.0",
     "libsodium-wrappers": "^0.8.4",