192 lines
7.0 KiB
TypeScript
192 lines
7.0 KiB
TypeScript
export type FileScopeClassificationReason =
|
|
| "included-write-scope"
|
|
| "invalid-entry"
|
|
| "duplicate-entry"
|
|
| "read-only-context"
|
|
| "forbidden-or-non-goal"
|
|
| "wrong-worktree-safeguard"
|
|
| "route-or-action"
|
|
| "fusion-metadata-evidence"
|
|
| "generated-lock"
|
|
| "conditional-changeset";
|
|
|
|
export interface FileScopeClassificationEntry {
|
|
token: string;
|
|
included: boolean;
|
|
reason: FileScopeClassificationReason;
|
|
line: string;
|
|
}
|
|
|
|
export interface FileScopeClassificationResult {
|
|
entries: FileScopeClassificationEntry[];
|
|
effectiveWriteScope: string[];
|
|
}
|
|
|
|
/*
|
|
FNXC:FileScopeClassification 2026-06-25-04:34:
|
|
Task File Scope is operator intent, not every path-like token in PROMPT.md. Keep this classifier conservative so read-only evidence, wrong-worktree safeguards, generated locks, route names, and conditional changesets do not create false write-scope leases or file-scope merge guards.
|
|
*/
|
|
const KNOWN_FILE_SCOPE_ROOT_FILES = new Set([
|
|
"makefile",
|
|
"dockerfile",
|
|
"justfile",
|
|
"license",
|
|
"readme",
|
|
"changelog",
|
|
"agents.md",
|
|
"project.yml",
|
|
"package.json",
|
|
"pnpm-lock.yaml",
|
|
]);
|
|
|
|
const INCLUDE_CONTEXT_RE = /\b(expected|touched|touch|modify|modified|write|writes|implementation|must update|artifacts?|files? changed|source paths?)\b/i;
|
|
const EXCLUDE_CONTEXT_RE = /\b(forbidden|non-goals?|out of scope|do not edit|do not modify|must not edit|must not modify|do not hand-edit|hand-edit|read-only|context to read|evidence only|metadata|wrong[- ]worktree|safeguards?)\b/i;
|
|
const GENERATED_CONTEXT_RE = /\b(generated|lockfiles?|locks?)\b/i;
|
|
const CONDITIONAL_CONTEXT_RE = /\b(conditional|only if|if .*changes|if .*changed|expected if|required only if|unless)\b/i;
|
|
const ROUTE_OR_ACTION_RE = /^(?:\/[A-Za-z0-9:_*?./-]+|fn_task_[A-Za-z0-9_]+|review|merge|retry|archive)$/i;
|
|
|
|
export function isValidFileScopeEntry(token: string): boolean {
|
|
const trimmed = token.trim();
|
|
if (!trimmed) return false;
|
|
|
|
const lower = trimmed.toLowerCase();
|
|
if (
|
|
lower.startsWith("origin/")
|
|
|| lower.startsWith("upstream/")
|
|
|| lower.startsWith("refs/")
|
|
|| /^https?:\/\//i.test(trimmed)
|
|
|| /^git@/i.test(trimmed)
|
|
|| /^ssh:\/\//i.test(trimmed)
|
|
|| /^[a-z]+\/fn-\d+$/i.test(trimmed)
|
|
|| /^[a-f0-9]{7,}$/i.test(trimmed)
|
|
|| trimmed.includes("..")
|
|
|| trimmed.startsWith("/")
|
|
) {
|
|
return false;
|
|
}
|
|
|
|
const segments = trimmed.split("/");
|
|
const lastSegment = segments[segments.length - 1] ?? "";
|
|
const hasSlash = trimmed.includes("/");
|
|
const hasDotInLastSegment = lastSegment.includes(".");
|
|
|
|
if (KNOWN_FILE_SCOPE_ROOT_FILES.has(lastSegment.toLowerCase())) {
|
|
return true;
|
|
}
|
|
|
|
if (trimmed.includes("**") || trimmed.endsWith("/*") || (lastSegment.includes("*") && hasDotInLastSegment)) {
|
|
return true;
|
|
}
|
|
|
|
if (hasSlash && hasDotInLastSegment) {
|
|
return true;
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
export function extractFileScopeTokens(content: string): string[] {
|
|
const section = extractFileScopeSection(content);
|
|
if (!section) return [];
|
|
return extractBacktickedTokens(section);
|
|
}
|
|
|
|
export function extractEffectiveWriteScopeFromPrompt(content: string): string[] {
|
|
return classifyFileScopeFromPrompt(content).effectiveWriteScope;
|
|
}
|
|
|
|
export function classifyFileScopeFromPrompt(content: string): FileScopeClassificationResult {
|
|
/*
|
|
FNXC:FileScopeClassification 2026-06-25-04:34:
|
|
Context headings inside `## File Scope` change the meaning of backticked tokens. The state machine is line-oriented on purpose: execution specs often mix write targets with forbidden paths and evidence-only metadata in the same section.
|
|
*/
|
|
const section = extractFileScopeSection(content);
|
|
if (!section) return { entries: [], effectiveWriteScope: [] };
|
|
|
|
const entries: FileScopeClassificationEntry[] = [];
|
|
const effectiveWriteScope: string[] = [];
|
|
const seen = new Set<string>();
|
|
let context: "include" | "exclude" | "conditional" = "include";
|
|
|
|
for (const rawLine of section.split("\n")) {
|
|
const line = rawLine.trim();
|
|
if (!line) continue;
|
|
if (INCLUDE_CONTEXT_RE.test(line) && !EXCLUDE_CONTEXT_RE.test(line) && !CONDITIONAL_CONTEXT_RE.test(line)) {
|
|
context = "include";
|
|
}
|
|
if (EXCLUDE_CONTEXT_RE.test(line)) {
|
|
context = "exclude";
|
|
}
|
|
if (CONDITIONAL_CONTEXT_RE.test(line)) {
|
|
context = "conditional";
|
|
}
|
|
|
|
const tokens = extractBacktickedTokens(line);
|
|
for (const rawToken of tokens) {
|
|
const token = rawToken.trim();
|
|
const reason = classifyToken(token, line, context);
|
|
if (reason !== "included-write-scope") {
|
|
entries.push({ token, included: false, reason, line });
|
|
continue;
|
|
}
|
|
if (seen.has(token)) {
|
|
entries.push({ token, included: false, reason: "duplicate-entry", line });
|
|
continue;
|
|
}
|
|
seen.add(token);
|
|
effectiveWriteScope.push(token);
|
|
entries.push({ token, included: true, reason, line });
|
|
}
|
|
}
|
|
|
|
return { entries, effectiveWriteScope };
|
|
}
|
|
|
|
function classifyToken(
|
|
token: string,
|
|
line: string,
|
|
context: "include" | "exclude" | "conditional",
|
|
): FileScopeClassificationReason {
|
|
/*
|
|
FNXC:FileScopeClassification 2026-06-25-04:34:
|
|
Classification reasons must be stable enough for diagnostics while the include/exclude decision stays binary. Preserve specific exclusion reasons after validation so review/spec gates explain why a token was ignored instead of silently shrinking File Scope.
|
|
*/
|
|
if (ROUTE_OR_ACTION_RE.test(token)) return "route-or-action";
|
|
if (!isValidFileScopeEntry(token)) return "invalid-entry";
|
|
|
|
const lowerToken = token.toLowerCase();
|
|
const lowerLine = line.toLowerCase();
|
|
if (lowerToken.startsWith(".fusion/") || lowerToken === ".fusion") return "fusion-metadata-evidence";
|
|
if (/^packages\/[^/]+\/package\.resolved$/i.test(token) || /^packages\/\*\/package\.resolved$/i.test(token)) {
|
|
return "generated-lock";
|
|
}
|
|
if (lowerToken.startsWith(".changeset/") && (context === "conditional" || CONDITIONAL_CONTEXT_RE.test(line))) {
|
|
return "conditional-changeset";
|
|
}
|
|
if (context === "conditional") return "read-only-context";
|
|
if (context === "exclude") {
|
|
if (lowerLine.includes("wrong-worktree") || lowerLine.includes("wrong worktree") || lowerLine.includes("safeguard")) {
|
|
return "wrong-worktree-safeguard";
|
|
}
|
|
if (lowerLine.includes("forbidden") || lowerLine.includes("non-goal") || lowerLine.includes("do not edit") || lowerLine.includes("do not modify")) {
|
|
return "forbidden-or-non-goal";
|
|
}
|
|
return "read-only-context";
|
|
}
|
|
if (GENERATED_CONTEXT_RE.test(line) && lowerToken.endsWith("package.resolved")) return "generated-lock";
|
|
return "included-write-scope";
|
|
}
|
|
|
|
function extractFileScopeSection(content: string): string | null {
|
|
const headingMatch = content.match(/^##\s+File\s+Scope\s*$/m);
|
|
if (!headingMatch) return null;
|
|
const startIdx = headingMatch.index! + headingMatch[0].length;
|
|
const rest = content.slice(startIdx);
|
|
const nextHeading = rest.search(/\n##?\s/);
|
|
return nextHeading === -1 ? rest : rest.slice(0, nextHeading);
|
|
}
|
|
|
|
function extractBacktickedTokens(text: string): string[] {
|
|
return Array.from(text.matchAll(/`([^`]+)`/g), (match) => match[1]?.trim() ?? "").filter(Boolean);
|
|
}
|