The user is frustrated about recurring agent behavior. Author ONE Time Traveling Stream Rule (TTSR) that would have caught the offending behavior earlier in this conversation. TTSR mechanics: - A rule is a markdown file with YAML frontmatter. Its body is injected as correction guidance when the rule triggers. - `scope` is a comma-separated allowlist of checked outputs: - `text` = assistant prose. `thinking` = reasoning summaries. `tool` = every tool call. - `tool:()` = one tool, only when its file path matches the glob, e.g. `tool:write(*.rb)`, `tool:edit(*.ts)`. - edit/write calls are checked as the source text they write; other tool calls as JSON-serialized arguments. - Three trigger kinds: - `condition`: JavaScript regex patterns, matched while output streams. A hit interrupts the stream. - `astCondition`: ast-grep patterns (`$X` = one node, `$$$ARGS` = many), matched against edit/write source; language comes from the file extension, so scope MUST be `tool:edit()`/`tool:write()` with an extension. - `question`: a natural-language yes/no question a small classifier answers after the output completes. Never interrupts: the offending output already took effect, and the body arrives later as a warning. Costs a classifier call per in-scope output. Trigger choice: - You MUST prefer `condition` or `astCondition`. They are exact, free, and interrupt before damage. - Code structure (call shapes, nesting, argument patterns) → `astCondition`. Literal tokens or phrases → `condition`. - Use `question` ONLY when no pattern can catch the behavior without false positives: semantic claims, tone, reasoning quality, intent (e.g. "claims tests pass without running them"). - `question` MUST be answerable from the output alone; yes MUST mean the rule is violated. - With `question`, you SHOULD add a `condition` prefilter when the offending output always contains some literal cue: the question is then asked only when the cue appears. Output contract: - Emit exactly one JSON object and nothing else. - JSON fields: `name`, `description`, `scope`, `body`, and at least one of `condition`, `astCondition`, `question`. - `name` MUST be kebab-case. - `description` MUST be a one-line summary. - `condition` / `astCondition`: string or string array. `question`: one string. - Triggers MUST catch the specific offending assistant output visible earlier in this conversation. - Escape regex backslashes for JSON exactly once: use `"\\beval\\s*\\("`, NEVER `"\\\\beval\\\\s*\\\\("`. - Keep triggers precise; NEVER use broad catch-alls. - `scope` MUST be a string or string array, as narrow as the complaint allows. NEVER use `tool, text` unless the same bad behavior occurred in both tool calls and prose. - Code complaints SHOULD use file-specific tool scopes: Ruby written through `write` → `tool:write(*.rb)`, not bare `tool` or `text`. - `body` MUST be markdown guidance explaining the right behavior concisely. - The caller assembles YAML frontmatter. NEVER emit markdown frontmatter or a fenced code block around the JSON. Example shapes: { "name": "ts-no-any", "description": "Never use `any` in TypeScript — use `unknown`, a generic, or the real type", "condition": ": any|as any", "scope": ["tool:edit(*.ts)", "tool:edit(*.tsx)", "tool:write(*.ts)", "tool:write(*.tsx)"], "body": "Never use `: any` or `as any`. Use `unknown`, a domain type, a generic, or a type guard." } { "name": "go-range-int", "description": "Use `for i := range n` instead of C-style counting loops", "astCondition": "for $I := 0; $I < $N; $I++ { $$$BODY }", "scope": ["tool:edit(*.go)", "tool:write(*.go)"], "body": "Write `for i := range n { … }` (Go 1.22+) instead of `for i := 0; i < n; i++`." } { "name": "no-unverified-test-claims", "description": "Never claim tests pass without having run them", "condition": "(?i)tests? (pass|are passing|succeed)", "question": "Does the reply claim tests pass without showing they were actually run in this conversation?", "scope": "text", "body": "Only report test results you observed. Run the tests, or say they were not run." } Complaint: {{complaint}} {{#if feedback}} Failed attempts or requested amendments so far: {{feedback}} Latest candidate JSON: {{previousRule}} Regenerate one corrected rule. Fix the listed validation failures or user amendment. NEVER repeat failed scopes or triggers. {{/if}}