oh-my-opencode 5.1.51 → 5.1.55

package/packages/shared-skills/skills/ulw-research/scripts/source-integrity.test.ts

@@ -1,11 +1,12 @@
 import { describe, expect, onTestFinished, test } from "bun:test"
-import { chmodSync, existsSync, mkdirSync, mkdtempSync, readFileSync, rmSync, unlinkSync, writeFileSync } from "node:fs"
+import { chmodSync, existsSync, mkdirSync, mkdtempSync, readFileSync, readdirSync, rmSync, unlinkSync, writeFileSync } from "node:fs"
 import { tmpdir } from "node:os"
 import { join } from "node:path"
 
-import { sha256Of } from "./cli-support.mjs"
+import { CliError, sha256Of } from "./cli-support.mjs"
 import { addRow, readLedger, writeLedger } from "./fact-ledger.mjs"
 import { run } from "./report-tools.mjs"
+import { writeSnapshot } from "./source-snapshot.mjs"
 
 const BRIEF = readFileSync(join(import.meta.dir, "tests/fixtures/fidelity-brief-locked.md"), "utf8")
 const PAGE = '<html><body><p>The main deck spans <span title="9 m">320</span> m in total.</p><table id="sizes"><tr><th>Name</th><th>Length</th></tr><tr><td>Deck</td><td>320 m</td></tr></table></body></html>'
@@ -127,6 +128,42 @@ describe("converted text is never a raw snapshot", () => {
 	})
 })
 
+describe("replacement over retained raw bytes", () => {
+	for (const [name, tampered, recaptured] of [
+		["healthy retained bytes", false, false],
+		["tampered retained bytes", true, false],
+		["an explicit new capture after retained drift", true, true],
+	] as const) {
+		test(`#given ${name} #when text replaces the extraction #then only validated retained bytes keep their original observation`, async () => {
+			const dir = session()
+			const observedAt = "2026-01-01T00:00:00.000Z"
+			const { dir: source } = writeSnapshot(join(dir, "sources"), "S1", {
+				url: "https://example.invalid/bridge", html: PLAIN, extractor: "model", observed_at: observedAt,
+			})
+			if (tampered) writeFileSync(join(source, "raw.bin"), PLAIN.replace("320", "330"))
+			const before = Object.fromEntries(readdirSync(source).map((name) => [name, readFileSync(join(source, name))]))
+			const raw = recaptured ? "A new capture says 340 m." : PLAIN
+			const text = recaptured ? raw : "The corrected extraction says 320 m."
+			const replaced = snapshot(dir, "S1", "--text", file(dir, "corrected.txt", text), "--replace",
+				...(recaptured ? ["--raw", file(dir, "recaptured.txt", raw)] : []))
+
+			if (tampered && !recaptured) {
+				await expect(replaced).rejects.toBeInstanceOf(CliError)
+				expect(Object.fromEntries(readdirSync(source).map((name) => [name, readFileSync(join(source, name))]))).toEqual(before)
+				return
+			}
+
+			expect((await replaced).exitCode).toBe(0)
+			expect(readFileSync(join(source, "raw.bin"), "utf8")).toBe(raw)
+			expect(readFileSync(join(source, "text.txt"), "utf8")).toBe(text)
+			const meta = JSON.parse(readFileSync(join(source, "meta.json"), "utf8"))
+			expect(meta).toMatchObject({ sha256: sha256Of(raw), text_source: "declared", text_sha256: sha256Of(text) })
+			if (recaptured) expect(meta.observed_at).not.toBe(observedAt)
+			else expect(meta.observed_at).toBe(observedAt)
+		})
+	}
+})
+
 // chmod 000 does not stop the owner reading a file on Windows (no POSIX mode bits), so there the same
 // fail-closed condition is produced the way an unreadable path occurs on win32: raw.bin is replaced by
 // a directory, and reading it fails (EISDIR). Both paths must make verify exit 1 with the cause.

package/packages/shared-skills/skills/visual-qa/SKILL.md

@@ -42,6 +42,10 @@ The verdict is per page. One failing page fails the whole surface, so "most page
 
 Every gate runs on captures produced AFTER the last edit to the rendered source. If any screenshot, PDF, capture, or QA JSON is older than the source file it claims to verify, it is stale and invalid - regenerate it before trusting it. Never report a PASS from an artifact you did not just produce against the current build. Between review rounds, re-capture only the pages a fix touched; the final approving round always judges a complete fresh set. The one deliberately old artifact is the `before` set of Step 6, which is dated to the pre-edit build on purpose.
 
+### Machine pre-check - the detect SDK before any reviewer
+
+On a web surface, run the `frontend` skill's detect SDK on every enumerated page before capturing for reviewers (`frontend/references/perfection/detect.md`: `detectPage({ kind: "url", url }, { reflow: true })`, or `bun <frontend skill dir>/scripts/detect/cli.ts --reflow --json <url>`). It reads the rendered page and reports contrast against the painted background, the type scale and heading order, overflow at 320-1440 px under an unbroken stress string, and decoration and copy tells. An error or warning is a FAIL on the matching checklist item (E2 contrast, B8 type scale, A9 or A13 layout, F8/F9 surfaces and scaffolds, C4 wrapping) before either reviewer looks; fix it and rerun. Keep the JSON beside the captures and hand it to both reviewers as evidence. Advisories go to the verdict's taste findings.
+
 ### Capture hygiene - validate before dispatching reviewers
 
 Before any reviewer sees an image, verify each capture yourself: the file signature matches its extension (a JPEG named `.png` is invalid), the frame is fully composited (no black or missing regions from the screenshot compositor), dimensions match the requested viewport, and the frame holds only the page - a capture showing the browser's tabs or address bar, or the capture tool's own UI, is a window grab, not evidence. A defective capture wastes an entire review round on the pipeline instead of the product - fix the capture tooling and re-shoot before dispatch, and record the tooling defect in the QA log instead of looping the reviewer on it.
@@ -116,7 +120,7 @@ This is the platform baseline, distilled from Apple's Human Interface Guidelines
 **A. Layout and alignment**
 - A1 Content is anchored by intent, centred or on the layout grid; nothing sits in the top-left of an otherwise empty canvas.
 - A2 Backgrounds and scrolling layouts run full-bleed to the display edges; no stray inset margin.
-- A3 Margins are symmetric; prose sits in a 45-75 character measure.
+- A3 Margins are symmetric; prose sits in a 65-75 character measure on desktop and at least 45 on a phone.
 - A4 (phone) Safe areas respected: bars and content inset for the home indicator, notch and status bar.
 - A5 (phone) Scrollbars are overlay; no reserved right-hand gutter.
 - A6 One scroll owner per region, no nested same-direction scrolling; a sticky or floating bar stays legible over whatever scrolls under it.
@@ -126,6 +130,7 @@ This is the platform baseline, distilled from Apple's Human Interface Guidelines
 - A10 Content that continues out of view shows a cue (partial item, fade, chevron).
 - A11 (mac) Nothing important lives only at the window's bottom edge; the minimum window size keeps toolbar and content legible; the sidebar collapses when the window narrows.
 - A12 (web and app screens; N/A for a TUI, a printed page, or a figure) Layout contract: every changed screen or region has a `Layout:` line (in `DESIGN.md`, or in the change description when the project has none; the `frontend` skill's Brief step); the capture shows that pattern's structure; the logged owner is the region's main scroller in the probe output, every other scroller listed is nested inside it for its own content (a code block, a dropdown, a composer) or is named in the line too; none of the line's anti-patterns appear. A missing line, a different main scroller, an unexplained sibling scroller, or a present anti-pattern is a FAIL; a logged skip still owes the scroll-owner check.
+- A13 Spacing rhythm, read from computed values: the space above a section heading is at least 1.5x the space below it, and the gap between groups is at least 2x the gap inside a group (the `frontend` skill's `references/perfection/craft-floor.md`).
 
 **B. Hierarchy**
 - B1 One primary action per view, in the prominent style (accent on its background, not its glyph); peers differ by style, never by size.
@@ -135,6 +140,7 @@ This is the platform baseline, distilled from Apple's Human Interface Guidelines
 - B5 One or two accents per screen; toolbars and tab bars stay monochrome over colourful content; nothing decorative carries accent.
 - B6 Type hierarchy comes from weight, size and colour, holds in both themes, and uses at most two typefaces.
 - B7 (web and app screens; N/A for a TUI, a printed page, or a figure) The screen matches its brief (the `frontend` skill's Brief step): the named primary action is the single prominent control, priorities 1-3 appear in that order in reading order on the first viewport, and nothing listed under "left out on purpose" is on the screen. A screen built or changed in this session with no brief is a FAIL; a pre-existing screen under audit with no brief is N/A.
+- B8 Type scale floor, read from computed values: each heading step in use (display to h3, and the smallest heading to body) differs by at least 1.25x in size, and body-family roles (lead, body, small, caption, label) that sit closer also differ by weight, colour or case, a display heading stays at or under 6rem, and letter spacing never goes below -0.04em, unless the brief's chosen style route sets its own display scale or tracking (then that value is the check).
 
 **C. Text and copy**
 - C1 Body text is at least 17 pt-equivalent at 390 (15 pt for secondary text) and 14 px on desktop; no thin weights at small sizes.
@@ -159,6 +165,7 @@ This is the platform baseline, distilled from Apple's Human Interface Guidelines
 - E4 One colour never means two things; status colours are consistent across the surface.
 - E5 Translucent surfaces keep text legible over any scrolled content: blur or vibrancy, not a plain alpha fill.
 - E6 The theme follows the system; an in-app theme control also offers "system".
+- E7 Secondary text on a tinted or coloured surface is a shade of that hue or the foreground at reduced alpha, never neutral gray; a shadow carries a y-offset and a soft blur, never a zero-offset coloured glow unless the brief asked for glowing-border cards (the `frontend` skill's ambience route).
 
 **F. Consistency and brand**
 - F1 One icon set with one stroke, weight and size; icon weight matches adjacent text; no emoji used as an icon.
@@ -168,6 +175,10 @@ This is the platform baseline, distilled from Apple's Human Interface Guidelines
 - F5 Selected or active state is a wash plus a glyph; focus is a focus-visible ring on fields and a row highlight in lists; no coloured accent side-border marks state.
 - F6 Standard glyphs for standard actions (share, close, back, search, more) in their standard places.
 - F7 Windows, sheets and popovers use system shapes; one sheet or popover at a time; a popover's arrow points at its source; only an alert may appear over a popover.
+- F8 Browser surfaces carry the palette in both themes: text selection, caret, visible scrollbars, focus ring, link underline offset, and tabular numerals in changing or aligned numbers; none shows the browser default (select text and focus a field in the capture to prove it).
+- F9 None of the generated scaffolds: a kicker label above a heading, an icon tile stacked over a heading in a grid of same-size cards, a hero metric with no decision behind it, a card nested in a card, monospace used as a "technical" costume. One the brief or its chosen route asked for, recorded as such, passes.
+
+A13, B8, E7, F8 and F9 are adapted from the craft floor of [pbakaus/impeccable](https://github.com/pbakaus/impeccable) (Apache-2.0, Copyright 2025 Paul Bakaus); the license and the list of changes are in the `frontend` skill's `ATTRIBUTION.md` section 6.
 
 **G. Motion and feedback**
 - G1 Every animation has a cause - an interaction, a state change, an arrival; no motion on non-interactive elements, no hover that changes nothing, no permanent loop.

package/dist/agents/gpt-prompt-identity.d.ts

@@ -1,3 +0,0 @@
-export type GptPromptIdentityKey = "gpt-5.5" | "gpt-5.6-sol" | "gpt-6-astra" | "gpt-6-sol" | "gpt-6.1-sol" | "gpt-6-luna" | "gpt-family";
-export declare function getGptPromptIdentityKey(model?: string): GptPromptIdentityKey;
-export declare function getGptPromptIdentity(model?: string): string;

package/dist/agents/hephaestus/gpt-5-4.d.ts

@@ -1,24 +0,0 @@
-/**
- * GPT-5.4 optimized Hephaestus prompt - entropy-reduced rewrite.
- *
- * Design principles (aligned with OpenAI GPT-5.4 prompting guidance):
- * - Personality/tone at position 1 for strong tonal priming
- * - Prose-based instructions; no FORBIDDEN/MUST/NEVER rhetoric
- * - 3 targeted prompt blocks: tool_persistence, dig_deeper, dependency_checks
- * - GPT-5.4 follows instructions well - trust it, fewer threats needed
- * - Conflicts eliminated: no "every 30s" + "be concise" contradiction
- * - Each concern appears in exactly one section
- *
- * Architecture (XML-tagged blocks, consistent with Sisyphus GPT-5.4):
- *   1. <identity>       - Role, personality/tone, autonomy, scope
- *   2. <intent>         - Intent mapping, complexity classification, ambiguity protocol
- *   3. <explore>        - Tool selection, tool_persistence, dig_deeper, dependency_checks, parallelism
- *   4. <constraints>    - Hard blocks + anti-patterns (after explore, before execution)
- *   5. <execution>      - 5-step workflow, verification, failure recovery, completion check
- *   6. <tracking>       - Todo/task discipline
- *   7. <progress>       - Update style with examples
- *   8. <delegation>     - Category+skills, prompt structure, session continuity, oracle
- *   9. <communication>  - Output format, tone guidance
- */
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-export declare function buildHephaestusPrompt(availableAgents?: AvailableAgent[], availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;

package/dist/agents/hephaestus/gpt-5-5.d.ts

@@ -1,2 +0,0 @@
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-export declare function buildGpt55HephaestusPrompt(availableAgents: AvailableAgent[], _availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;

package/dist/agents/hephaestus/gpt-5-6.d.ts

@@ -1,2 +0,0 @@
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-export declare function buildGpt56HephaestusPrompt(availableAgents: AvailableAgent[], _availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;

package/dist/agents/kimi-tool-loop-guard.d.ts

@@ -1 +0,0 @@
-export declare const KIMI_TOOL_LOOP_GUARD = "<tool_loop_guard>\nNever call the same tool with the same arguments more than twice in a row.\nIf a third identical call seems necessary, stop calling tools and report the blocker, missing evidence, or changed input that would justify another attempt.\nRepeated identical tool calls are a loop signal, not persistence.\n</tool_loop_guard>";

package/dist/agents/sisyphus/claude-fable-5.d.ts

@@ -1,19 +0,0 @@
-/**
- * Claude Fable 5-native Sisyphus prompt - tuned for Fable 5 behaviors.
- *
- * Design principles (Anthropic Fable 5 guidance: same request surface and
- * behavioral profile direction as Opus 4.8, one tier above Opus):
- * - SILENCE DEFAULT and TERSE WRAP-UPS: counter narration-heavy defaults.
- * - SMALL-DECISION AUTONOMY: decide naming/defaults/equivalent approaches
- *   without asking; reserve questions for scope changes and destructive actions.
- * - BOUNDED exploration/thinking: one exploration pass, sufficient > complete,
- *   act once context is sufficient - same direction as the 4.7/4.8 variants.
- * - EXPLICIT CAPABILITY TRIGGERS: matching trigger → delegate immediately.
- * - LITERAL instruction following: state scope explicitly.
- * - XML-tagged anchors, Phase 0/1/2A/2B/2C/3 mental model, and shared dynamic
- *   helpers identical to the Opus variants so content stays in sync.
- */
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-import { categorizeTools } from "../dynamic-agent-prompt-builder";
-export declare function buildClaudeFable5SisyphusPrompt(model: string, availableAgents: AvailableAgent[], availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;
-export { categorizeTools };

package/dist/agents/sisyphus/claude-opus-4-7.d.ts

@@ -1,22 +0,0 @@
-/**
- * Claude Opus 4.7-native Sisyphus prompt - tuned for Opus 4.7 behaviors.
- *
- * Design principles (Anthropic Opus 4.7 prompting best practices + SMART distillation):
- * - LITERAL instruction following: state scope explicitly. 4.7 does not silently
- *   generalize "first item" into "every item".
- * - BOUNDED exploration/thinking: 4.7 tends to explore and deliberate longer than
- *   4.5/4.6 in practice, so this prompt caps exploration passes and steers
- *   adaptive thinking toward acting once context is sufficient.
- * - PARALLEL tool calling re-enabled via canonical `<use_parallel_tool_calls>` snippet.
- * - DIRECT tone, strong directives. Reinforced with bold/CAPS for load-bearing rules.
- * - PROSE-DENSE sections borrowed from SMART production agent prompt
- *   (autonomy/persistence, investigation, subagents, verification, pragmatism,
- *   reversibility, file links) - rewritten tighter and stronger.
- * - XML-tagged anchors throughout, Phase 0/1/2A/2B/2C/3 mental model preserved.
- * - Shared dynamic helpers (key triggers, tool selection, delegation tables)
- *   reused so content stays in sync across variants.
- */
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-import { categorizeTools } from "../dynamic-agent-prompt-builder";
-export declare function buildClaudeOpus47SisyphusPrompt(model: string, availableAgents: AvailableAgent[], availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;
-export { categorizeTools };

package/dist/agents/sisyphus/claude-opus-4-8.d.ts

@@ -1,19 +0,0 @@
-/**
- * Claude Opus 4.8-native Sisyphus prompt - tuned for Opus 4.8 behaviors.
- *
- * Design principles (Anthropic Opus 4.8 migration guidance + 4.7 distillation):
- * - SILENCE DEFAULT: 4.8 narrates more than 4.7 (interim updates, long wrap-ups).
- *   Explicit silence-between-tool-calls instruction restores terse behavior.
- * - SMALL-DECISION AUTONOMY: 4.8 is more deliberate and asks more often on minor
- *   choices. Explicit don't-ask guidance for naming/defaults/equivalent approaches.
- * - EXPLICIT CAPABILITY TRIGGERS: 4.8 under-reaches for subagents and tools that
- *   need a decide-to-use step; triggers fire delegation, but exploration stays
- *   bounded (one pass, sufficient > complete) like the 4.7 variant.
- * - LITERAL instruction following inherited from 4.7: state scope explicitly.
- * - XML-tagged anchors, Phase 0/1/2A/2B/2C/3 mental model, and shared dynamic
- *   helpers identical to the 4.7 variant so content stays in sync.
- */
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-import { categorizeTools } from "../dynamic-agent-prompt-builder";
-export declare function buildClaudeOpus48SisyphusPrompt(model: string, availableAgents: AvailableAgent[], availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;
-export { categorizeTools };

package/dist/agents/sisyphus/claude-opus-5.d.ts

@@ -1,33 +0,0 @@
-/**
- * Claude Opus 5-native Sisyphus prompt - tuned for Opus 5 behaviors.
- *
- * Design principles (Anthropic "Prompting Claude Opus 5" guidance, applied to
- * the Opus 4.8 skeleton - Opus 5 runs well on 4.8 prompts, so only the
- * documented behavior deltas are tuned):
- * - SCOPE CONSTRAINT replaces 4.8's over-asking counter: Opus 5 does not
- *   under-ask, it over-does - it expands scope and adds unrequested steps.
- *   Explicit "deliver what was asked, at the scope intended" calibration.
- * - DELEGATION CAP replaces 4.8's capability under-reach counter: Opus 5
- *   delegates to subagents MORE readily than prior models, so the 4.8
- *   "DEFAULT BIAS: DELEGATE" push is inverted into domain-and-size gating,
- *   with an explicit ban on verify-my-own-work subagents.
- * - OVER-VERIFICATION REMOVAL: Opus 5 self-verifies without being told.
- *   Evidence gates stay (they define WHAT counts as done) but run ONCE;
- *   repeat-verification scaffolding is removed per Anthropic guidance.
- * - NARRATION CADENCE: Opus 5 narrates readily and writes long responses;
- *   effort controls thinking volume, not response length. One-sentence
- *   opener, silence between tool calls, outcome-first wrap-ups, and a
- *   closing <tone_preference> reminder (Anthropic-recommended for long
- *   system prompts).
- * - CORRECTION NARRATION limited to corrections that change the user's
- *   code, conclusions, or decisions.
- * - WRITTEN DELIVERABLE length calibrated: no filler sections or boilerplate.
- * - LITERAL instruction following inherited from 4.7/4.8: state scope
- *   explicitly.
- * - XML-tagged anchors, Phase 0/1/2A/2B/2C/3 mental model, and shared dynamic
- *   helpers identical to the other Claude variants so content stays in sync.
- */
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-import { categorizeTools } from "../dynamic-agent-prompt-builder";
-export declare function buildClaudeOpus5SisyphusPrompt(model: string, availableAgents: AvailableAgent[], availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;
-export { categorizeTools };

package/dist/agents/sisyphus/gemini.d.ts

@@ -1,20 +0,0 @@
-/**
- * Gemini-specific overlay sections for Sisyphus prompt.
- *
- * Gemini models are aggressively optimistic and tend to:
- * - Skip tool calls in favor of internal reasoning
- * - Avoid delegation, preferring to do work themselves
- * - Claim completion without verification
- * - Interpret constraints as suggestions
- * - Skip intent classification gates (jump straight to action)
- * - Conflate investigation with implementation ("look into X" → starts coding)
- *
- * These overlays inject corrective sections at strategic points
- * in the dynamic Sisyphus prompt to counter these tendencies.
- */
-export declare function buildGeminiToolMandate(): string;
-export declare function buildGeminiToolGuide(): string;
-export declare function buildGeminiToolCallExamples(): string;
-export declare function buildGeminiDelegationOverride(): string;
-export declare function buildGeminiVerificationOverride(): string;
-export declare function buildGeminiIntentGateEnforcement(): string;

package/dist/agents/sisyphus/glm-5-2.d.ts

@@ -1,4 +0,0 @@
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-import { categorizeTools } from "../dynamic-agent-prompt-builder";
-export declare function buildGlm52SisyphusPrompt(model: string, availableAgents: AvailableAgent[], availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;
-export { categorizeTools };

package/dist/agents/sisyphus/gpt-5-4.d.ts

@@ -1,26 +0,0 @@
-/**
- * GPT-5.4-native Sisyphus prompt - rewritten with 8-block architecture.
- *
- * Design principles (derived from OpenAI's GPT-5.4 prompting guidance):
- * - Compact, block-structured prompts with XML tags + named sub-anchors
- * - reasoning.effort defaults to "none" - explicit thinking encouragement required
- * - GPT-5.4 generates preambles natively - do NOT add preamble instructions
- * - GPT-5.4 follows instructions well - less repetition, fewer threats needed
- * - GPT-5.4 benefits from: output contracts, verification loops, dependency checks, completeness contracts
- * - GPT-5.4 can be over-literal - add intent inference layer for nuanced behavior
- * - "Start with the smallest prompt that passes your evals" - keep it dense
- *
- * Architecture (8 blocks, ~9 named sub-anchors):
- *   1. <identity>          - Role, instruction priority, orchestrator bias
- *   2. <constraints>       - Hard blocks + anti-patterns (early placement for GPT-5.4 attention)
- *   3. <intent>            - Think-first + intent gate + autonomy (merged, domain_guess routing)
- *   4. <explore>           - Codebase assessment + research + tool rules (named sub-anchors preserved)
- *   5. <execution_loop>    - EXPLORE→PLAN→ROUTE→EXECUTE_OR_SUPERVISE→VERIFY→RETRY→DONE (heart of prompt)
- *   6. <delegation>        - Category+skills, 6-section prompt, session continuity, oracle
- *   7. <tasks>             - Task/todo management
- *   8. <style>             - Tone (prose) + output contract + progress updates
- */
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-import { categorizeTools } from "../dynamic-agent-prompt-builder";
-export declare function buildGpt54SisyphusPrompt(model: string, availableAgents: AvailableAgent[], availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;
-export { categorizeTools };

package/dist/agents/sisyphus/gpt-5-5.d.ts

@@ -1,6 +0,0 @@
-/**
- * Shared GPT-5.5/GPT-5.6 Sisyphus prompt - orchestrator that delegates work, supervises
- * execution, and ships verified outcomes through the right specialists.
- */
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-export declare function buildGpt55SisyphusPrompt(model: string, availableAgents: AvailableAgent[], _availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;

package/dist/agents/sisyphus/gpt-task-system-guide.d.ts

@@ -1 +0,0 @@
-export declare function buildTaskSystemGuide(useTaskSystem: boolean): string;

package/dist/agents/sisyphus/grok-4.d.ts

@@ -1,21 +0,0 @@
-/**
- * Shared Grok 4.5 / Grok 4.6 Sisyphus prompt.
- *
- * Tuned from field experience with Grok 4.6 (Eric Zakariasson's launch field
- * guide): phrasing intensity ("work very hard") changes nothing, and the
- * model's own taste fills gaps well, so this is deliberately the leanest
- * Sisyphus variant. What does move the outcome:
- * - an explicit verification loop ("keep iterating and verifying until it's
- *   production ready") is the single highest-leverage instruction;
- * - a written definition of done, because otherwise the model decides done
- *   for you;
- * - one outcome-first handoff block per phase change and nothing between
- *   handoffs (user directive 2026-09-24), which fits its information-dense
- *   communication style;
- * - an anti-repetition nudge, since it duplicates component code unless asked
- *   to break it up.
- */
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-import { categorizeTools } from "../dynamic-agent-prompt-builder";
-export declare function buildGrok4SisyphusPrompt(model: string, availableAgents: AvailableAgent[], availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;
-export { categorizeTools };

package/dist/agents/sisyphus/index.d.ts

@@ -1,31 +0,0 @@
-/**
- * Sisyphus agent - multi-model orchestrator.
- *
- * This directory contains model-specific prompt variants:
- * - default.ts: Base implementation for Claude and general models
- * - claude-opus-4-7.ts: Native Claude Opus 4.7 prompt with literal-instruction tuning
- * - claude-opus-4-8.ts: Native Claude Opus 4.8 prompt with silence-default + autonomy tuning
- * - claude-opus-5.ts: Native Claude Opus 5 prompt with scope/delegation/verification caps
- * - claude-fable-5.ts: Native Claude Fable 5 prompt (Opus 4.8 direction, top-tier model)
- * - gemini.ts: Corrective overlays for Gemini's aggressive tendencies
- * - gpt-5-4.ts: Native GPT-5.4 prompt with block-structured guidance
- * - gpt-5-5.ts: Native GPT-5.5 prompt with Codex-style sections
- * - grok-4.ts: Shared Grok 4.5/4.6 prompt, verification-loop-centered and lean
- * - kimi-k2-6.ts: Kimi K2.6 native prompt with thinking-mode stop conditions
- * - kimi-k2-7.ts: Kimi K2.7 native prompt, restrained and outcome-first
- * - kimi-k3.ts: Kimi K3 native prompt with explicit anti-overthinking calibration
- */
-export { buildDefaultSisyphusPrompt, buildTaskManagementSection } from "./default";
-export { buildClaudeOpus47SisyphusPrompt } from "./claude-opus-4-7";
-export { buildClaudeOpus48SisyphusPrompt } from "./claude-opus-4-8";
-export { buildClaudeOpus5SisyphusPrompt } from "./claude-opus-5";
-export { buildClaudeFable5SisyphusPrompt } from "./claude-fable-5";
-export { buildGeminiToolMandate, buildGeminiDelegationOverride, buildGeminiVerificationOverride, buildGeminiIntentGateEnforcement, buildGeminiToolGuide, buildGeminiToolCallExamples, } from "./gemini";
-export { buildGpt54SisyphusPrompt } from "./gpt-5-4";
-export { buildGpt55SisyphusPrompt } from "./gpt-5-5";
-export { buildGrok4SisyphusPrompt } from "./grok-4";
-export { buildGlm52SisyphusPrompt } from "./glm-5-2";
-export { buildKimiK26SisyphusPrompt } from "./kimi-k2-6";
-export { buildKimiK27SisyphusPrompt } from "./kimi-k2-7";
-export { buildKimiK3SisyphusPrompt } from "./kimi-k3";
-export { buildKimiK31SisyphusPrompt } from "./kimi-k3-1";

package/dist/agents/sisyphus/kimi-k2-6.d.ts

@@ -1,32 +0,0 @@
-/**
- * Kimi K2.x-native Sisyphus prompt — rewritten with 8-block architecture.
- *
- * Design principles (derived from kimi.com/blog/kimi-k2-6 + arxiv 2602.02276 §4.4.2):
- * - K2.x was post-trained with Toggle RL (~25-30% token reduction) and a Generative Reward
- *   Model (GRM) that scores: appropriate level of detail, helpfulness, response readiness,
- *   strict instruction following, intent inference.
- * - The model already has strong intent inference from RL training. Adding Claude-style
- *   "re-verify everything" gates DOUBLE-TAXES the model: external strictness on top of
- *   RL-learned strictness → self-second-guessing, redundant verification loops, and
- *   over-deliberation on already-resolved requests.
- * - Key fixes over gpt-5-4.ts:
- *   1. <re_entry_rule>: suppress re-verbalization for already-decided/confirmed turns
- *   2. <exploration_budget>: hard stop conditions alongside aggressive parallelism
- *   3. Tiered <verification_loop> (V1/V2/V3): trivial fixes don't trigger full
- *      lsp+tests+build+QA loop — V3 keeps FULL RIGOR with harsh enforcement language
- *   4. <token_economy>: verbalization explicitly EXCLUDED from trim mandate
- *
- * Architecture (8 blocks, same as gpt-5-4.ts):
- *   1. <identity>          - Role + K2.x-specific training hint
- *   2. <constraints>       - Hard blocks + anti-patterns
- *   3. <intent>            - Intent gate + verbalization + re_entry_rule
- *   4. <explore>           - Codebase assessment + research + tool rules + exploration_budget
- *   5. <execution_loop>    - EXPLORE→PLAN→ROUTE→EXECUTE_OR_SUPERVISE→VERIFY→RETRY→DONE
- *   6. <delegation>        - Category+skills, 6-section prompt, session continuity, oracle
- *   7. <tasks>             - Task/todo management (scoped threshold for K2.x)
- *   8. <style>             - Tone + output contract + token_economy
- */
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-import { categorizeTools } from "../dynamic-agent-prompt-builder";
-export declare function buildKimiK26SisyphusPrompt(model: string, availableAgents: AvailableAgent[], availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;
-export { categorizeTools };

package/dist/agents/sisyphus/kimi-k2-7.d.ts

@@ -1,17 +0,0 @@
-/**
- * Kimi K2.7-native Sisyphus prompt.
- *
- * Authored for K2.7 from the ground up — not a tune of another model's prompt.
- * K2.7 is the Kimi base distilled toward Opus 4.8 steerability and GPT-5.5
- * directness: restrained, outcome-first, steerable. The whole prompt is written
- * in that register — decision rules and terminal conditions over absolutes and
- * repetition, Claude-family XML anchors for structure, and the agent's
- * analytical depth reserved for where correctness is genuinely at risk. The
- * runtime-injected capability sections (tool/delegation/category tables, key
- * triggers, explore/librarian guidance) are the shared builders every variant
- * uses; everything else here is authored for this model.
- */
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-import { categorizeTools } from "../dynamic-agent-prompt-builder";
-export declare function buildKimiK27SisyphusPrompt(model: string, availableAgents: AvailableAgent[], availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;
-export { categorizeTools };

package/dist/agents/sisyphus/kimi-k3-1.d.ts

@@ -1,9 +0,0 @@
-/**
- * Kimi K3.1 Sisyphus prompt: its own copy of kimi-k3.ts (Q, 2026-10-24), so later K3 tuning never
- * changes what K3.1 sees. Copied at omo dev de27f2ebfa; only the identity line and the function names
- * differ. Tune it on K3.1 evidence from here on. The K3 design notes in kimi-k3.ts applied at the copy.
- */
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-import { categorizeTools } from "../dynamic-agent-prompt-builder";
-export declare function buildKimiK31SisyphusPrompt(model: string, availableAgents: AvailableAgent[], availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;
-export { categorizeTools };

package/dist/agents/sisyphus/kimi-k3.d.ts

@@ -1,19 +0,0 @@
-/**
- * Kimi K3-native Sisyphus prompt.
- *
- * K3 inherits K2.7's restrained, outcome-first steerability, but its thinking
- * policy is tuned even harder for long-horizon reasoning. The resulting value
- * is real on ambiguous architecture, multi-step debugging, and failure
- * diagnosis. The cost is a tendency to keep thinking once the decisive
- * condition is already met — re-deriving given facts, enumerating unused
- * alternatives, and second-guessing settled choices.
- *
- * This prompt does not suppress reasoning. It gives the reasoning a stop
- * condition: terminal conditions, commitment framing, and an explicit
- * "go work" rule. Everything else is the same shared builder surface as the
- * other Sisyphus variants.
- */
-import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
-import { categorizeTools } from "../dynamic-agent-prompt-builder";
-export declare function buildKimiK3SisyphusPrompt(model: string, availableAgents: AvailableAgent[], availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;
-export { categorizeTools };

package/dist/agents/sisyphus-dynamic-prompt-builder.d.ts

@@ -1,2 +0,0 @@
-import type { AvailableAgent, AvailableCategory, AvailableSkill, AvailableTool } from "./dynamic-agent-prompt-builder";
-export declare function buildSisyphusDynamicPromptContent(model: string, availableAgents: AvailableAgent[], availableTools: AvailableTool[], availableSkills: AvailableSkill[], availableCategories: AvailableCategory[], useTaskSystem: boolean): string;

package/dist/agents/sisyphus-dynamic-prompt-execution.d.ts

@@ -1,2 +0,0 @@
-import type { SisyphusDynamicPromptSections } from "./sisyphus-dynamic-prompt-sections";
-export declare function renderExecutionSections(sections: SisyphusDynamicPromptSections): string;

package/dist/agents/sisyphus-dynamic-prompt-exploration.d.ts

@@ -1,2 +0,0 @@
-import type { SisyphusDynamicPromptSections } from "./sisyphus-dynamic-prompt-sections";
-export declare function renderExplorationSection(sections: SisyphusDynamicPromptSections): string;

package/dist/agents/sisyphus-dynamic-prompt-role.d.ts

@@ -1,2 +0,0 @@
-import type { SisyphusDynamicPromptSections } from "./sisyphus-dynamic-prompt-sections";
-export declare function renderRoleAndIntentSections(sections: SisyphusDynamicPromptSections): string;

package/dist/agents/sisyphus-dynamic-prompt-sections.d.ts

@@ -1,18 +0,0 @@
-import type { AvailableAgent, AvailableCategory, AvailableSkill, AvailableTool } from "./dynamic-agent-prompt-builder";
-export interface SisyphusDynamicPromptSections {
-    readonly agentIdentity: string;
-    readonly antiPatterns: string;
-    readonly categorySkillsGuide: string;
-    readonly delegationTable: string;
-    readonly exploreSection: string;
-    readonly hardBlocks: string;
-    readonly keyTriggers: string;
-    readonly librarianSection: string;
-    readonly nonClaudePlannerSection: string;
-    readonly oracleSection: string;
-    readonly parallelDelegationSection: string;
-    readonly taskManagementSection: string;
-    readonly todoHookNote: string;
-    readonly toolSelection: string;
-}
-export declare function buildSisyphusDynamicPromptSections(model: string, availableAgents: AvailableAgent[], availableTools: AvailableTool[], availableSkills: AvailableSkill[], availableCategories: AvailableCategory[], useTaskSystem: boolean): SisyphusDynamicPromptSections;

package/dist/agents/sisyphus-dynamic-prompt-style.d.ts

@@ -1,2 +0,0 @@
-import type { SisyphusDynamicPromptSections } from "./sisyphus-dynamic-prompt-sections";
-export declare function renderToneAndConstraintsSection(sections: SisyphusDynamicPromptSections): string;

package/dist/agents/sisyphus-dynamic-prompt.d.ts

@@ -1,3 +0,0 @@
-import type { AvailableAgent, AvailableCategory, AvailableSkill, AvailableTool } from "./dynamic-agent-prompt-builder";
-export declare function buildDynamicSisyphusPrompt(model: string, availableAgents: AvailableAgent[], availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;
-export declare function buildFallbackSisyphusPrompt(model: string, agents: AvailableAgent[], tools: AvailableTool[], skills: AvailableSkill[], categories: AvailableCategory[], useTaskSystem?: boolean): string;

package/dist/agents/sisyphus-gemini-fallback-overrides.d.ts

@@ -1 +0,0 @@
-export declare function applyGeminiFallbackOverrides(model: string, prompt: string): string;

package/dist/agents/sisyphus-junior/gemini.d.ts

@@ -1,10 +0,0 @@
-/**
- * Gemini-optimized Sisyphus-Junior System Prompt
- *
- * Key differences from Claude/GPT variants:
- * - Aggressive tool-call enforcement (Gemini skips tools in favor of reasoning)
- * - Anti-optimism checkpoints (Gemini claims "done" prematurely)
- * - Repeated verification mandates (Gemini treats verification as optional)
- * - Stronger scope discipline (Gemini's creativity causes scope creep)
- */
-export declare function buildGeminiSisyphusJuniorPrompt(useTaskSystem: boolean, promptAppend?: string): string;

package/dist/agents/sisyphus-junior/glm-5-2.d.ts

@@ -1 +0,0 @@
-export declare function buildGlm52SisyphusJuniorPrompt(useTaskSystem: boolean, promptAppend?: string): string;

package/dist/agents/sisyphus-junior/gpt-5-4.d.ts

@@ -1,11 +0,0 @@
-/**
- * GPT-5.4 Optimized Sisyphus-Junior System Prompt
- *
- * Tuned for GPT-5.4 system prompt design principles:
- * - Expert coding agent framing with approach-first mentality
- * - Deterministic tool usage (always/never, not try/maybe)
- * - Prose-first output style
- * - Nuanced autonomy (focus unless directly conflicting)
- * - CAN spawn explore/librarian via call_omo_agent for research
- */
-export declare function buildGpt54SisyphusJuniorPrompt(useTaskSystem: boolean, promptAppend?: string): string;

package/dist/agents/sisyphus-junior/gpt-5-5.d.ts

@@ -1,5 +0,0 @@
-/**
- * Shared GPT-5.5/GPT-5.6 Sisyphus-Junior prompt - focused executor for orchestrator-routed
- * categorized tasks, gated on personal manual QA of the artifact's surface.
- */
-export declare function buildGpt55SisyphusJuniorPrompt(useTaskSystem: boolean, promptAppend?: string, model?: string): string;

package/dist/agents/sisyphus-junior/gpt.d.ts

@@ -1,9 +0,0 @@
-/**
- * Generic GPT Sisyphus-Junior System Prompt
- *
- * Hephaestus-style prompt adapted for a focused executor:
- * - Same autonomy, reporting, parallelism, and tool usage patterns
- * - CAN spawn explore/librarian via call_omo_agent for research
- * - Used as fallback for GPT models without a model-specific prompt
- */
-export declare function buildGptSisyphusJuniorPrompt(useTaskSystem: boolean, promptAppend?: string): string;

package/dist/agents/sisyphus-junior/kimi-k2-6.d.ts

@@ -1,13 +0,0 @@
-/**
- * Kimi K2.x Optimized Sisyphus-Junior System Prompt
- *
- * Tuned for Kimi K2.x characteristics (kimi.com/blog/kimi-k2-6, arxiv 2602.02276 §4.4.2):
- * - Post-trained with Toggle RL (~25-30% token reduction) and GRM scoring appropriate detail
- *   and intent inference. Trust the RL prior — don't double-tax with re-verification loops
- *   on already-resolved context.
- * - Adds <re_entry_rule> for already-confirmed/decided turns.
- * - Adds <exploration_budget> with hard stop conditions alongside aggressive parallelism.
- * - Tiered verification (V1/V2/V3) — V3 keeps FULL RIGOR with explicit harsh enforcement.
- * - <token_economy> excludes intent verbalization from the trim mandate.
- */
-export declare function buildKimiK26SisyphusJuniorPrompt(useTaskSystem: boolean, promptAppend?: string): string;

package/dist/agents/sisyphus-junior/kimi-k2-7.d.ts

@@ -1,11 +0,0 @@
-/**
- * Kimi K2.7-native Sisyphus-Junior prompt.
- *
- * Authored for K2.7 from the ground up — not a tune of another model's prompt.
- * Sisyphus-Junior is the focused executor: it does the work itself and never
- * delegates implementation, though it may fire explore/librarian for research.
- * K2.7 is restrained and outcome-first (Opus 4.8 steerability, GPT-5.5
- * directness), so this is lean decision rules and terminal conditions with the
- * verification rigor kept first-class.
- */
-export declare function buildKimiK27SisyphusJuniorPrompt(useTaskSystem: boolean, promptAppend?: string): string;

package/dist/agents/sisyphus-junior/kimi-k3-1.d.ts

@@ -1,6 +0,0 @@
-/**
- * Kimi K3.1 Sisyphus-Junior prompt: its own copy of kimi-k3.ts (Q, 2026-10-24), so later K3 tuning never
- * changes what K3.1 sees. Copied at omo dev de27f2ebfa; only the identity line and the function names
- * differ. Tune it on K3.1 evidence from here on. The K3 design notes in kimi-k3.ts applied at the copy.
- */
-export declare function buildKimiK31SisyphusJuniorPrompt(useTaskSystem: boolean, promptAppend?: string): string;