Files
waggle-os/packages/agent/src/skill-redaction.ts
shumeiko.it 2e1999a3f6 refactor: remove self-learning and the optimizer package (P1.4)
Self-learning is excluded from the community edition (ADR-0004); the
project owner reworks it separately. Cost tracking stays.

- agent: delete behaviour evolution (evolve-schema, iterative GEPA,
  compose-evolution, evolution gates, orchestrator and deploy),
  improvement and correction detection, AgentLearning, the eval-dataset
  builder, optimization-capture and the agent_insights tool. Secret
  detection moves from eval-dataset to secret-patterns. The LLM wiring
  keeps only the Haiku adapter and judge call used by skill audit and
  hygiene, so @ax-llm/ax is now declared by agent instead of server.
- server: delete /api/evolution/*, the evolution and optimizer services,
  GEPA expansion of the first chat message, the prompt_optimization and
  monthly_assessment jobs, the idle-session reviewer and its
  session-reviewer persona, prompt injection of past corrections and
  thumbs-down recorded as correction signals. Existing job rows of the
  removed types are ignored, not deleted.
- core: delete OptimizationLogStore; stop re-exporting the improvement
  signal and evolution run stores.
- web: delete the Evolution screen and the two job types in the
  automation editor.
- delete packages/optimizer.

Also logs UF-027 (self-learning loops read data nothing collects) and
UF-028 (the Settings prompt-shape choice is ignored by chat).
2026-10-20 00:54:58 +02:00

63 lines
2.2 KiB
TypeScript

/**
* Pre-write redaction for authored skill content.
*
* The closed-learning-loop (create_skill / auto_extract_skills) and the UI both
* persist skill markdown to disk (~/.waggle/skills). That content can carry
* secrets (API keys / tokens) or machine-specific filesystem paths pulled from
* the model's context or the user's input. The loop's "strip secrets/paths"
* instruction was model-advisory only — this enforces it deterministically
* before every authored write.
*
* Secrets reuse the curated SECRET_PATTERNS via redactSecrets (single source of
* truth — see secret-patterns.ts). Paths cover the user's home directory (the exact
* running-user path plus generic per-OS forms), which both leaks the username
* and makes a skill non-portable.
*/
import os from 'node:os';
import { redactSecrets } from './secret-patterns.js';
// Generic user-home prefixes. Each matches up to (not including) the next path
// separator, so the tail is preserved: C:\Users\me\.waggle\x -> ~\.waggle\x
const HOME_PATH_PATTERNS: readonly RegExp[] = [
/[A-Za-z]:\\Users\\[^\\/:*?"<>|\r\n]+/g, // Windows
/\/(?:Users|home)\/[^\s/\\:*?"<>|]+/g, // macOS / Linux
];
export interface SkillRedactionResult {
content: string;
/** De-duplicated labels of what was stripped (secret pattern names + 'home-path'). */
redactions: string[];
}
/**
* Strip secrets and user-home paths from skill content before it is written to
* disk. Pure — returns a new string; never mutates input.
*/
export function redactSkillContent(content: string): SkillRedactionResult {
const sec = redactSecrets(content);
let out = sec.text;
const redactions = new Set(sec.found);
// Exact running-user home first (most precise; literal string match).
try {
const home = os.homedir();
if (home && out.includes(home)) {
out = out.split(home).join('~');
redactions.add('home-path');
}
} catch {
/* homedir unavailable — generic patterns below still apply */
}
// Generic home paths (content authored about other machines/users).
for (const re of HOME_PATH_PATTERNS) {
const replaced = out.replace(re, '~');
if (replaced !== out) {
out = replaced;
redactions.add('home-path');
}
}
return { content: out, redactions: [...redactions] };
}