mirror of
https://github.com/garrytan/gstack.git
synced 2026-11-20 15:13:47 +01:00
voice: reply in the language of the user's latest message
Every skill's voice directive now tells the agent to reply in the language of the user's latest message unless asked otherwise, so long English skill text no longer drags a Chinese (or any non-English) session into English. Mixed-language sessions follow the latest message. Code, commands, paths, identifiers and quoted output stay verbatim, and at tier 2+ so do the machine-read question markers (D<N>, option letters, (recommended)). The rule costs 205 bytes per tier-2+ skill and 144 per tier-1 skill. A further 25-byte cut to the bounded closer's exemption list leaves the net for this wave at +17 bytes per tier-2+ skill and +158 per tier-1 skill. review, qa and browse caps rise with measured notes; the ship goldens are refreshed. Co-authored-by: Xianying Tan <shrektan@126.com> Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
This commit is contained in:
1 parent
8d1db315af
commit
e5dc011b7c
56 files changed
+138
-47
No files matched your search
+2
-1
@@ -222,12 +222,13 @@ GStack voice: Garry-shaped product and engineering judgment.
|
||||
- Sound like a builder talking to a builder, not a consultant presenting to a client.
|
||||
- Never corporate, academic, PR, or hype. Avoid filler, throat-clearing, generic optimism, and founder cosplay.
|
||||
- No em dashes. No AI vocabulary: delve, crucial, robust, comprehensive, nuanced, multifaceted, furthermore, moreover, additionally, pivotal, landscape, tapestry, underscore, foster, showcase, intricate, vibrant, fundamental, significant, load-bearing.
|
||||
- Reply in the language of the user's latest message unless asked otherwise. Code, commands, paths, identifiers, quoted output and question markers (`D<N>`, option letters, `(recommended)`) stay verbatim.
|
||||
- The user has context you do not: domain knowledge, timing, relationships, taste. Cross-model agreement is a recommendation, not a decision. The user decides.
|
||||
|
||||
Good: "auth.ts:47 returns undefined when the session cookie expires. Users hit a white screen. Fix: add a null check and redirect to /login. Two lines."
|
||||
Bad: "I've identified a potential issue in the authentication flow that may cause problems under certain conditions."
|
||||
|
||||
**Bounded closer.** After completing work, report in at most a few short lines: what changed, what was skipped, what to watch. No feature tours or unrequested design notes. Exempt: AskUserQuestion decision briefs, completion-status blocks, explanations the user asked for, and a skill's mandated report (/qa-only, /plan-*-review, /retro, /document-generate). The rule limits prose around the deliverable, never the deliverable.
|
||||
**Bounded closer.** After completing work, report in at most a few short lines: what changed, what was skipped, what to watch. No feature tours or unrequested design notes. Exempt: decision briefs, completion-status blocks, requested explanations, and a skill's mandated report (/qa-only, /plan-*-review, /retro, /document-generate). The rule limits prose around the deliverable, never the deliverable.
|
||||
|
||||
Good closer: "Renamed the flag in 3 files, regenerated docs, tests green. Skipped the CLI alias (unused since v1.2); watch the Windows job."
|
||||
Bad closer: a tour of every edit, a restatement of the plan, and three paragraphs justifying choices nobody questioned.
|
||||
|
||||
+2
-1
@@ -225,12 +225,13 @@ GStack voice: Garry-shaped product and engineering judgment.
|
||||
- Sound like a builder talking to a builder, not a consultant presenting to a client.
|
||||
- Never corporate, academic, PR, or hype. Avoid filler, throat-clearing, generic optimism, and founder cosplay.
|
||||
- No em dashes. No AI vocabulary: delve, crucial, robust, comprehensive, nuanced, multifaceted, furthermore, moreover, additionally, pivotal, landscape, tapestry, underscore, foster, showcase, intricate, vibrant, fundamental, significant, load-bearing.
|
||||
- Reply in the language of the user's latest message unless asked otherwise. Code, commands, paths, identifiers, quoted output and question markers (`D<N>`, option letters, `(recommended)`) stay verbatim.
|
||||
- The user has context you do not: domain knowledge, timing, relationships, taste. Cross-model agreement is a recommendation, not a decision. The user decides.
|
||||
|
||||
Good: "auth.ts:47 returns undefined when the session cookie expires. Users hit a white screen. Fix: add a null check and redirect to /login. Two lines."
|
||||
Bad: "I've identified a potential issue in the authentication flow that may cause problems under certain conditions."
|
||||
|
||||
**Bounded closer.** After completing work, report in at most a few short lines: what changed, what was skipped, what to watch. No feature tours or unrequested design notes. Exempt: AskUserQuestion decision briefs, completion-status blocks, explanations the user asked for, and a skill's mandated report (/qa-only, /plan-*-review, /retro, /document-generate). The rule limits prose around the deliverable, never the deliverable.
|
||||
**Bounded closer.** After completing work, report in at most a few short lines: what changed, what was skipped, what to watch. No feature tours or unrequested design notes. Exempt: decision briefs, completion-status blocks, requested explanations, and a skill's mandated report (/qa-only, /plan-*-review, /retro, /document-generate). The rule limits prose around the deliverable, never the deliverable.
|
||||
|
||||
Good closer: "Renamed the flag in 3 files, regenerated docs, tests green. Skipped the CLI alias (unused since v1.2); watch the Windows job."
|
||||
Bad closer: a tour of every edit, a restatement of the plan, and three paragraphs justifying choices nobody questioned.
|
||||
|
||||
+2
-1
@@ -206,12 +206,13 @@ GStack voice: Garry-shaped product and engineering judgment.
|
||||
- Sound like a builder talking to a builder, not a consultant presenting to a client.
|
||||
- Never corporate, academic, PR, or hype. Avoid filler, throat-clearing, generic optimism, and founder cosplay.
|
||||
- No em dashes. No AI vocabulary: delve, crucial, robust, comprehensive, nuanced, multifaceted, furthermore, moreover, additionally, pivotal, landscape, tapestry, underscore, foster, showcase, intricate, vibrant, fundamental, significant, load-bearing.
|
||||
- Reply in the language of the user's latest message unless asked otherwise. Code, commands, paths, identifiers, quoted output and question markers (`D<N>`, option letters, `(recommended)`) stay verbatim.
|
||||
- The user has context you do not: domain knowledge, timing, relationships, taste. Cross-model agreement is a recommendation, not a decision. The user decides.
|
||||
|
||||
Good: "auth.ts:47 returns undefined when the session cookie expires. Users hit a white screen. Fix: add a null check and redirect to /login. Two lines."
|
||||
Bad: "I've identified a potential issue in the authentication flow that may cause problems under certain conditions."
|
||||
|
||||
**Bounded closer.** After completing work, report in at most a few short lines: what changed, what was skipped, what to watch. No feature tours or unrequested design notes. Exempt: AskUserQuestion decision briefs, completion-status blocks, explanations the user asked for, and a skill's mandated report (/qa-only, /plan-*-review, /retro, /document-generate). The rule limits prose around the deliverable, never the deliverable.
|
||||
**Bounded closer.** After completing work, report in at most a few short lines: what changed, what was skipped, what to watch. No feature tours or unrequested design notes. Exempt: decision briefs, completion-status blocks, requested explanations, and a skill's mandated report (/qa-only, /plan-*-review, /retro, /document-generate). The rule limits prose around the deliverable, never the deliverable.
|
||||
|
||||
Good closer: "Renamed the flag in 3 files, regenerated docs, tests green. Skipped the CLI alias (unused since v1.2); watch the Windows job."
|
||||
Bad closer: a tour of every edit, a restatement of the plan, and three paragraphs justifying choices nobody questioned.
|
||||
|
||||
@@ -484,7 +484,7 @@ do not launch the downstream skill or open a browser.`,
|
||||
gateAfterStop: undefined, // operational multi-STOP skill, like ship
|
||||
},
|
||||
behavioral: 'plan',
|
||||
maxSkeletonBytes: 74_881, // Shared-code identity/skip/action rules + critical-severity validation; measured 74,493 (2026-09-17). + v1.91.12.0 merge of #2999 (review clarity repairs: await reads, research alongside dispatch, /review deadline and setup authority, findings sources) with #3002 (guarded state-root lines, plan-check checkpoints); each fit alone; measured 74,881 (2026-10-01).
|
||||
maxSkeletonBytes: 74_884, // Shared-code identity/skip/action rules + critical-severity validation; measured 74,493 (2026-09-17). + v1.91.12.0 merge of #2999 (review clarity repairs: await reads, research alongside dispatch, /review deadline and setup authority, findings sources) with #3002 (guarded state-root lines, plan-check checkpoints); each fit alone; measured 74,881 (2026-10-01). + Oct 8 wave: shared voice directive bans "load-bearing" and adds the reply-language rule, offset by tightening the bounded closer (net +17 bytes per tier-2+ skill); measured 74,884 (2026-10-08).
|
||||
minUnionBytes: 89_000, // Phase 4 wave 1; measured union 93,357
|
||||
mustContain: ['confidence', 'P1', 'P2', 'Review Army', 'adversarial'],
|
||||
maxSizeRatio: 1.2345, // Shared-code feature + critical-severity validation: 128,042 union bytes / 108,523 baseline = 1.1799; preserves content floors. + v1.91.12.0 merge of #2999 (above, plus plan-completion fallback intent and specialist checklist-by-path) with #3002; measured 1.1843 (2026-10-01). + v1.91.19.0 wave: verdict-form Codex callers with sandbox preflight (INV-1/B1), P0/P1 structured gate (B1b), bound-plan discovery and data-not-instructions guard (F1/D3), visible LEARNINGS failures (B5); measured 129,941 / 108,523 = 1.1974 (2026-10-03). + v1.91.18.0 (#3023) merged into the v1.91.19.0 wave; measured 130,025 / 108,523 = 1.1981 (2026-10-04). + Oct 6 fix wave: quota_exhausted preflight branch (#3051); measured 130,359 / 108,523 = 1.2012 (2026-10-06). + Oct 6 fix wave A3/CEO-12: free text reaches commands only through agent-written mktemp files (mktemp block, write rule, path rebuild and refusal at each posting, prompt and brief site); measured 131,314 / 108,523 = 1.2100 (2026-10-06). Combined Oct 6 wave head (A and B): measured 131,966 / 108,523 = 1.2160 (2026-10-06). + Oct 7 wave B3: the Codex probe runs as a command (select-model capture lines, check-sandbox/probe-model/run-with-timeout subcommands) instead of being sourced; measured 132,199 / 108,523 = 1.2182 (2026-10-07). + Oct 7 wave CEO-28/DX-11: review prompts ask for a severity per finding and design proposals use the proposal gate; measured 132,258 / 108,523 = 1.2187 (2026-10-07). + Oct 7 wave guard (CEO-20): conditional `run_in_background` request plus the background-run recovery at each foreground dispatch site; measured 132,233 / 108,523 = 1.2185 (2026-10-07). Merged with the guard area: measured 132,522 / 108,523 = 1.2211 (2026-10-07). + Oct 8 wave: the defensive-security framing and fixture summary mode reach the security specialist and the Red Team (stated once in the dispatch, referenced by the Red Team); measured 133,902 / 108,523 = 1.2339 (2026-10-08).
|
||||
@@ -670,7 +670,7 @@ do not launch the downstream skill or open a browser.`,
|
||||
},
|
||||
behavioral: 'prompt',
|
||||
maxSkeletonBytes: 63_500, // + v2.0 {{ASIDE_SETUP}}/{{BROWSE_FALLBACK}} (replaces the browse setup block); measured 61_253
|
||||
maxSizeRatio: 1.127, // + v1.81 Aside contract + gstack-browser fallback block (1.080 on v1.91.7.0) + the shared test value bar at 8a.5 ({{TEST_VALUE_BAR:qa}}); measured 1.094 + W1 guarded state-root resolution (`eval gstack-paths; : "${GSTACK_STATE_ROOT:?…}"`) in the Context Recovery preamble, the eureka log and each state-writing bash block; measured 1.101 (2026-09-30) + v1.91.12.0 merge of #2999 (await scope/method Reads, capture --after checkpoints, browser-only empty evidence list) with #3002; measured 1.1028 (2026-10-01). + both untrusted-content marker formats in the browser fallback (W7f); measured 1.1035 (2026-10-02). Re-measured 1.1038 after the W8 probe-loop renames (2026-10-03); the cap already matches. + the annotate-every-capture and superseded rule before materialize (a ship-exploratory-late-input run published evidence that omitted older-snapshot captures); measured 1.1064 (2026-10-03). + v1.91.18.0 (#3023) merged into the v1.91.19.0 wave; measured 93,282 / 84,049 = 1.1099 (2026-10-04). + reliability follow-ups C2 (judge-named gaps after trimming the added wording: feasible 180 s Browser Quick, PROBE_DIR's source line, stop goes to the §4 report, written JSON keeps full card values; lane S's {user}/timing fields); measured 93,311 / 84,049 = 1.11020 (2026-10-04). + Oct 6 fix wave A3/CEO-12: free text reaches commands only through agent-written mktemp files (mktemp block, write rule, path rebuild and refusal at each posting, prompt and brief site); measured 94,486 / 84,049 = 1.1242 (2026-10-06). + Oct 7 wave B6: Aside rules 1, 4, 6 and 7 (offer a target-origin tab before openTab, a second sign-in wall means a tab- or URL-bound session, no early-return hint, fast [ok without the sentinel is an abort); measured 94,681 / 84,049 = 1.1265 (2026-10-07).
|
||||
maxSizeRatio: 1.1272, // + v1.81 Aside contract + gstack-browser fallback block (1.080 on v1.91.7.0) + the shared test value bar at 8a.5 ({{TEST_VALUE_BAR:qa}}); measured 1.094 + W1 guarded state-root resolution (`eval gstack-paths; : "${GSTACK_STATE_ROOT:?…}"`) in the Context Recovery preamble, the eureka log and each state-writing bash block; measured 1.101 (2026-09-30) + v1.91.12.0 merge of #2999 (await scope/method Reads, capture --after checkpoints, browser-only empty evidence list) with #3002; measured 1.1028 (2026-10-01). + both untrusted-content marker formats in the browser fallback (W7f); measured 1.1035 (2026-10-02). Re-measured 1.1038 after the W8 probe-loop renames (2026-10-03); the cap already matches. + the annotate-every-capture and superseded rule before materialize (a ship-exploratory-late-input run published evidence that omitted older-snapshot captures); measured 1.1064 (2026-10-03). + v1.91.18.0 (#3023) merged into the v1.91.19.0 wave; measured 93,282 / 84,049 = 1.1099 (2026-10-04). + reliability follow-ups C2 (judge-named gaps after trimming the added wording: feasible 180 s Browser Quick, PROBE_DIR's source line, stop goes to the §4 report, written JSON keeps full card values; lane S's {user}/timing fields); measured 93,311 / 84,049 = 1.11020 (2026-10-04). + Oct 6 fix wave A3/CEO-12: free text reaches commands only through agent-written mktemp files (mktemp block, write rule, path rebuild and refusal at each posting, prompt and brief site); measured 94,486 / 84,049 = 1.1242 (2026-10-06). + Oct 7 wave B6: Aside rules 1, 4, 6 and 7 (offer a target-origin tab before openTab, a second sign-in wall means a tab- or URL-bound session, no early-return hint, fast [ok without the sentinel is an abort); measured 94,681 / 84,049 = 1.1265 (2026-10-07). + Oct 8 wave: shared voice directive bans "load-bearing" and adds the reply-language rule, offset by tightening the bounded closer (net +17 bytes per tier-2+ skill); measured 94,736 / 84,049 = 1.12715 (2026-10-08).
|
||||
minUnionBytes: 69_500, // measured union 70,385
|
||||
// 'aside repl' pins the Aside contract; '$B goto' pins the fallback block in the always-loaded skeleton.
|
||||
mustContain: ['bug', 'aside repl', '$B goto', 'fix', 'Health Score Rubric', 'regression'],
|
||||
@@ -707,7 +707,7 @@ do not launch the downstream skill or open a browser.`,
|
||||
gateAfterStop: undefined,
|
||||
},
|
||||
behavioral: 'prompt',
|
||||
maxSkeletonBytes: 34_100, // + v2.0 {{ASIDE_SETUP}}/{{BROWSE_FALLBACK}} skeleton + the STOP-Read pointer; measured 29_986 + Oct 6 fix wave A3/CEO-12: free text reaches commands only through agent-written mktemp files (mktemp block, write rule, path rebuild and refusal at each posting, prompt and brief site); skeleton measured 32,532 (2026-10-06). + Oct 7 wave B6: Aside rules 1, 4, 6 and 7 (offer a target-origin tab before openTab, a second sign-in wall means a tab- or URL-bound session, no early-return hint, fast [ok without the sentinel is an abort), and the cookbook's silent-failure list and signed-in-tab recipe (the rule details live here, not in every browsing skill); skeleton measured 34,047 (2026-10-07).
|
||||
maxSkeletonBytes: 34_210, // + v2.0 {{ASIDE_SETUP}}/{{BROWSE_FALLBACK}} skeleton + the STOP-Read pointer; measured 29_986 + Oct 6 fix wave A3/CEO-12: free text reaches commands only through agent-written mktemp files (mktemp block, write rule, path rebuild and refusal at each posting, prompt and brief site); skeleton measured 32,532 (2026-10-06). + Oct 7 wave B6: Aside rules 1, 4, 6 and 7 (offer a target-origin tab before openTab, a second sign-in wall means a tab- or URL-bound session, no early-return hint, fast [ok without the sentinel is an abort), and the cookbook's silent-failure list and signed-in-tab recipe (the rule details live here, not in every browsing skill); skeleton measured 34,047 (2026-10-07). + Oct 8 wave: the tier-1 voice directive bans "load-bearing" and adds the reply-language rule (tier 1 has no closer to tighten: +158 bytes); skeleton measured 34,209 (2026-10-08).
|
||||
minUnionBytes: 36_000, // skeleton + command-list section (14,240); measured union ~38.5 KB
|
||||
// 'BEGIN/END UNTRUSTED EXTERNAL' pins the untrusted-content warning in the
|
||||
// command reference; 'aside repl' pins the Aside contract in the skeleton.
|
||||
|
||||
@@ -33,6 +33,8 @@ const LATER_WAVE_EDITS: Array<[string, string]> = [
|
||||
// Oct 7 wave: the artifacts-sync preamble block surfaces attention lines (one line longer).
|
||||
["Skill-start already ran artifacts sync. GBrain hint text (if any) says\nwhen to prefer `gbrain` over Grep. `ARTIFACTS_SYNC:` reports sync health\n(`off`, `mode=... | queue=N`, `remote-mode`, or a `gstack-brain-restore`\nhint). On an `attention:` line, tell the user in one sentence what\nit says and the command it names, then continue.\n\nThe one-time privacy stop-gate arrives as a `GSTACK_INSTRUCTION` block\nfrom skill-start when consent is pending; fire it via AskUserQuestion\nexactly as instructed.",
|
||||
"The skill-start output above already ran artifacts sync. Act on its lines:\nGBrain hint text (if present) tells you when to prefer `gbrain` over Grep;\n`ARTIFACTS_SYNC:` reports sync health (`off`, `mode=... | queue=N`,\n`remote-mode`, or a restore hint naming `gstack-brain-restore`).\n\nThe one-time privacy stop-gate (artifacts-sync consent) arrives as a\n`GSTACK_INSTRUCTION` block from skill-start when consent is actually pending\n\u2014 fire it via AskUserQuestion exactly as the block instructs."],
|
||||
// Oct 8 wave: the tier-1 voice directive gained the reply-language rule (two lines longer).
|
||||
["Reply in the language of the user's latest message unless asked otherwise. Code, commands, paths, identifiers and quoted output stay verbatim.\n\n", ''],
|
||||
];
|
||||
|
||||
export function approvedCookieWorkflowSource(source: string): string {
|
||||
|
||||
@@ -1,6 +1,10 @@
|
||||
import { describe, expect, test } from 'bun:test';
|
||||
import { execFileSync } from 'node:child_process';
|
||||
import { readFileSync } from 'node:fs';
|
||||
import { join } from 'node:path';
|
||||
import { generateVoiceDirective } from '../scripts/resolvers/preamble/generate-voice-directive';
|
||||
import { HOST_PATHS, type TemplateContext } from '../scripts/resolvers/types';
|
||||
import { between, expectMentions } from './helpers/prompt-structure';
|
||||
|
||||
const ctx: TemplateContext = { skillName: 'review', tmplPath: 'review/SKILL.md.tmpl', host: 'claude', paths: HOST_PATHS.claude };
|
||||
const tiers: number[] = [1, 3];
|
||||
@@ -11,4 +15,25 @@ describe('shared voice directive', () => {
|
||||
expect(banLine).toBeDefined();
|
||||
expect(banLine!).toContain('load-bearing');
|
||||
});
|
||||
|
||||
test.each(tiers)('tier %i replies in the language of the latest message and keeps technical text verbatim', tier => {
|
||||
const voice = generateVoiceDirective(ctx, tier);
|
||||
expectMentions(voice, [
|
||||
['reply', 'language', "user's latest message", 'unless'],
|
||||
['code', 'commands', 'paths', 'identifiers', 'quoted output', 'verbatim'],
|
||||
], `tier ${tier} voice`);
|
||||
if (tier >= 2) expectMentions(voice, [['question markers', '`D<N>`', 'option letters', '`(recommended)`', 'verbatim']], 'tier 3 voice');
|
||||
});
|
||||
|
||||
test('every generated skill with a Voice section carries the reply-language rule', () => {
|
||||
const root = join(import.meta.dir, '..');
|
||||
const skills = execFileSync('git', ['ls-files', 'SKILL.md', '*/SKILL.md'], { cwd: root, encoding: 'utf8', timeout: 10_000 }).split('\n').filter(Boolean);
|
||||
const voiced = skills.filter(file => readFileSync(join(root, file), 'utf8').includes('\n## Voice\n'));
|
||||
expect(voiced.length).toBeGreaterThan(40);
|
||||
const missing = voiced.filter(file => {
|
||||
const voice = between(readFileSync(join(root, file), 'utf8'), '\n## Voice\n', /\n## /).toLowerCase();
|
||||
return !(voice.includes('language') && voice.includes('latest message') && voice.includes('verbatim'));
|
||||
});
|
||||
expect(missing).toEqual([]);
|
||||
});
|
||||
});
|
||||
Reference in new issue
Block a user