skillcellar 0.3.0 → 0.6.1

package/src/secrets.mjs

@@ -0,0 +1,75 @@
+// What stops a backup: a line that looks like a key, a token or a password. Each pattern is bounded,
+// so a line of any length is read in one pass, and a placeholder such as ${NAME}, $NAME, <your key>
+// or your_api_key is not a secret. A finding is a reason to look, not proof: the backup stops either
+// way, and a key of a shape no pattern knows is not found.
+
+// A key stands alone: nothing alphanumeric right before or after it, so STRIPE_API_KEY= and "apiKey"
+// count, which a word boundary would miss between _ and a letter.
+const EDGE = String.raw`(?<![A-Za-z0-9])`;
+const END = String.raw`(?![A-Za-z0-9])`;
+// A value that is not a placeholder, a variable, a template, an example or a reference into config or code.
+const NOT_PLACEHOLDER = String.raw`(?!\$\{|\$[A-Za-z_]|<|\{\{|%|your[_-]|xxx|example|placeholder|change[_-]?me|redacted|none\b|null\b|true\b|false\b|\*{3}|process\.|os\.|env\.|config\.|settings\.|this\.|self\.)`;
+// The characters an assigned or bearer value is read with.
+const VALUE = String.raw`[A-Za-z0-9_\-./+=]`;
+// A word in either case, for a pattern that minds case.
+const anyCase = (word) => word.replace(/[a-z]/g, (letter) => `[${letter}${letter.toUpperCase()}]`);
+const EXAMPLE_WORDS = ["dummy", "fake", "sample", "example", "placeholder", "redacted", "change[_-]?me"].map(anyCase).join("|");
+// A value as documentation writes one, not a key: elided (abc123... or abc123…), or holding a word,
+// not part of a longer one, that calls it an example. Real keys are neither.
+const NOT_EXAMPLE = String.raw`(?!${VALUE}{0,255}?(?:\.\.\.|(?<![A-Za-z])(?:${EXAMPLE_WORDS})(?![A-Za-z])))(?!${VALUE}{1,255}…)`;
+// Code such as base64.b64encode( is no value; a quoted value is one, whatever it holds.
+const NOT_CALL = String.raw`(?:(?<=["'\x60])|(?![A-Za-z_]\w{0,60}(?:\.[A-Za-z_]\w{0,60}){1,5}\())`;
+
+const PATTERNS = [
+  ["a private key", /-----BEGIN (?:RSA |EC |DSA |OPENSSH |PGP |ENCRYPTED )?PRIVATE KEY(?: BLOCK)?-----/],
+  ["an AWS access key", new RegExp(String.raw`${EDGE}AKIA[0-9A-Z]{16}${END}`)],
+  ["a GitHub token", new RegExp(String.raw`${EDGE}(?:gh[pousr]_[A-Za-z0-9]{36,255}|github_pat_[A-Za-z0-9_]{22,255})${END}`)],
+  ["an OpenAI or Anthropic key", new RegExp(String.raw`${EDGE}sk-(?:proj-|ant-|svcacct-)?[A-Za-z0-9_-]{20,255}(?![A-Za-z0-9_-])`)],
+  ["a Slack token", new RegExp(String.raw`${EDGE}xox[abeprs]-[A-Za-z0-9-]{10,255}(?![A-Za-z0-9-])`)],
+  ["a Stripe key", new RegExp(String.raw`${EDGE}(?:[sr]k_(?:live|test)_[A-Za-z0-9]{16,255}|whsec_[A-Za-z0-9]{16,255})${END}`)],
+  ["a Google API key", new RegExp(String.raw`${EDGE}AIza[0-9A-Za-z_-]{35}(?![A-Za-z0-9_-])`)],
+  ["a SkillCellar API key", new RegExp(String.raw`${EDGE}skc_[A-Za-z0-9]{64}${END}`)],
+  ["an npm token", new RegExp(String.raw`${EDGE}npm_[A-Za-z0-9]{36}${END}`)],
+  ["a Neon API key", new RegExp(String.raw`${EDGE}napi_[a-z0-9]{40,255}${END}`)],
+  ["a Resend, Hugging Face or Groq key", new RegExp(String.raw`${EDGE}(?:re_[A-Za-z0-9_]{20,255}|hf_[A-Za-z0-9]{20,255}|gsk_[A-Za-z0-9]{20,255})${END}`)],
+  ["a JSON Web Token", new RegExp(String.raw`${EDGE}eyJ[A-Za-z0-9_-]{10,2000}\.[A-Za-z0-9_-]{10,2000}\.[A-Za-z0-9_-]{10,2000}(?![A-Za-z0-9_-])`)],
+  // user:password@host in an address of any scheme, unless the password is a placeholder.
+  ["a password in an address", /[a-z][a-z0-9+.-]{0,30}:\/\/[^\s/@:'"]{1,100}:(?!\$\{)[^\s/@'"]{1,200}@/i],
+  ["a bearer or basic credential", new RegExp(String.raw`\b(?:Bearer|Basic)\s{1,5}(?!\$\{|\$[A-Za-z_]|<)${NOT_EXAMPLE}[A-Za-z0-9_\-./+=]{16,2000}(?![A-Za-z0-9_\-./+=])`)],
+  // A key named as one, assigned a value with a digit in it, as random keys have; the name may be quoted, as in JSON.
+  [
+    "a key or password in an assignment",
+    new RegExp(
+      String.raw`${EDGE}(?:api[_-]?key|apikey|secret[_-]?(?:access[_-]?)?key|access[_-]?token|auth[_-]?token|refresh[_-]?token|client[_-]?secret|private[_-]?key|password|passwd|secret|token)${END}["'\x60]?\s{0,5}[:=]\s{0,5}["'\x60]?${NOT_PLACEHOLDER}${NOT_EXAMPLE}${NOT_CALL}(?=[A-Za-z0-9_\-./+=]{0,254}[0-9])[A-Za-z0-9_\-./+=]{8,255}`,
+      "i",
+    ),
+  ],
+];
+
+/**
+ * The first line of a text that looks like a secret, as `{ line, kind }` (the line counted from
+ * 1), or null. Lines are read one by one, so the file's size sets the cost, never a pattern.
+ */
+export function findSecret(text) {
+  const lines = text.split("\n");
+  for (let index = 0; index < lines.length; index += 1) {
+    const line = lines[index];
+    for (const [kind, pattern] of PATTERNS) if (pattern.test(line)) return { line: index + 1, kind };
+  }
+  return null;
+}
+
+/** The placeholders a sanitized MCP config holds, such as ${CONTEXT7_KEY}, in order and once each. */
+export function placeholdersIn(text) {
+  return [...new Set(Array.from(text.matchAll(/\$\{([A-Z0-9_]{1,80})\}/g), (match) => match[0]))];
+}
+
+/**
+ * References in an MCP config that Claude Code would expand from the environment and that are not
+ * plain placeholders: `${VAR:-default}`, `$VAR`, or a placeholder of another shape. Each is a reason
+ * to read the server's config before adding it, since it reaches into the user's environment.
+ */
+export function oddReferencesIn(text) {
+  const all = Array.from(text.matchAll(/\$\{[^}]{0,200}\}|\$[A-Za-z_][A-Za-z0-9_]{0,80}/g), (match) => match[0]);
+  return [...new Set(all.filter((reference) => !/^\$\{[A-Z0-9_]{1,80}\}$/.test(reference)))];
+}