@ultrakit/ultrakit 0.0.0

package/dist/template/.opencode/command/ui-slop-check.md

@@ -0,0 +1,169 @@
+---
+description: Audit changed UI files for AI slop patterns and design-system violations
+argument-hint: "[path|auto] [--staged] [--since=<ref>] [--full-report]"
+agent: vision
+model: proxypal/gemini-3-pro-preview
+---
+
+# UI Slop Check: $ARGUMENTS
+
+Run a focused anti-slop audit against changed UI files using the frontend-design taxonomy.
+
+## Load Skills
+
+```typescript
+skill({ name: "frontend-design" }); // Anti-pattern taxonomy + design references
+skill({ name: "visual-analysis" }); // Structured visual/code analysis workflow
+skill({ name: "accessibility-audit" }); // Keyboard/focus/contrast checks
+skill({ name: "ux-quality-gates" }); // UX correctness gates beyond visual slop
+```
+
+## Parse Arguments
+
+| Argument        | Default | Description                                                 |
+| --------------- | ------- | ----------------------------------------------------------- |
+| `[path\|auto]`  | `auto`  | Specific file/dir to audit, or auto-detect changed UI files |
+| `--staged`      | false   | Audit staged changes only (`git diff --cached`)             |
+| `--since=<ref>` | `HEAD`  | Compare against ref (`main`, `HEAD~1`, commit SHA)          |
+| `--full-report` | false   | Include all categories even when no issues found            |
+
+## Phase 1: Resolve Target Files
+
+If `[path]` is provided:
+
+- Audit that path directly
+
+If `auto`:
+
+```bash
+# unstaged + staged by default
+git diff --name-only $SINCE_REF -- \
+  '*.tsx' '*.jsx' '*.css' '*.scss' '*.sass' '*.less' '*.html' '*.mdx'
+```
+
+If `--staged`:
+
+```bash
+git diff --cached --name-only -- \
+  '*.tsx' '*.jsx' '*.css' '*.scss' '*.sass' '*.less' '*.html' '*.mdx'
+```
+
+Prioritize files under:
+
+- `src/components/**`
+- `src/app/**`
+- `src/pages/**`
+- `app/**`
+- `components/**`
+
+If no UI files changed, return: **PASS (no changed UI files)**.
+
+## Phase 2: Run AI Slop Checklist
+
+Evaluate each target file (or rendered screenshot if provided) against these checks.
+
+### A) Typography
+
+- Banned default aesthetics (Inter/Roboto/Arial/Open Sans as dominant display voice)
+- Body text uses `rem/em`, not fixed `px`
+- Clear hierarchy (size/weight/spacing), not color-only hierarchy
+- Body line length near readable measure (around 65ch when applicable)
+
+### B) Color and Theming
+
+- No AI default palette tropes (purple-blue gradient defaults, neon-on-dark clichés)
+- No pure `#000` / `#fff` as dominant surfaces
+- Gray text is not placed on saturated backgrounds
+- Semantic tokens are used (not random per-component hardcoded colors)
+- Dark mode is adapted, not simple inversion
+
+### C) Layout and Spatial Rhythm
+
+- No cards-inside-cards without strong information architecture reason
+- No repetitive cookie-cutter card blocks with identical structure
+- Spacing rhythm is consistent (4pt-style cadence), not arbitrary jumps
+- Uses `gap`/layout primitives cleanly; avoids margin hacks when possible
+
+### D) Motion and Interaction
+
+- No bounce/elastic gimmick motion for product UI
+- Animations use transform/opacity (avoid layout-thrashing properties)
+- Reduced motion support exists for meaningful motion
+- States exist: hover, focus-visible, active, disabled, loading/error where relevant
+
+### E) UX Writing
+
+- Buttons are verb + object (e.g. "Save changes")
+- Error copy includes what happened + why + how to fix
+- Empty states include guidance + next action
+- Terminology is consistent (avoid mixed synonyms for same action)
+- User-facing labels avoid implementation terms, database names, and internal acronyms
+
+### F) UX Quality Gates
+
+- One dominant filled primary action per view/section
+- Destructive actions require explicit confirm or undo, with specific entity/count in copy
+- No placeholder-as-label form fields
+- Form errors are associated with controls (`aria-describedby`, `aria-invalid`, `role="alert"`)
+- Submit/loading states prevent double-submit without layout shift
+- Empty, loading, error, and success states exist where async/data flows exist
+- Error toasts/banners persist long enough and include retry/undo/support where applicable
+- Data-heavy UI distinguishes empty state from filtered no-results state
+- Bulk actions show selected count and confirm destructive scope
+
+### G) Accessibility Safety Nets
+
+- Keyboard-visible focus treatment (`:focus-visible`)
+- Contrast baseline expectations (WCAG AA)
+- Touch targets reasonable (44x44 context where applicable)
+- Native semantic elements are used before ARIA patches (`button`, `a`, `form`, landmarks)
+- Heading structure has one logical `h1` per page/screen context
+
+### H) Component Family Consistency
+
+- Buttons and inputs in the same form share height, radius, border, and focus treatment
+- Focus, error, disabled, and loading states use the same token logic across components
+- No one-off radius/shadow/border width unless documented as a system-level exception
+- Semantic color roles are consistent: success, warning, destructive, info, primary
+
+## Phase 3: Severity and Scoring
+
+Group findings by severity:
+
+- **Critical**: accessibility failures, broken interaction states, unreadable contrast
+- **Warning**: strong AI fingerprint/slop patterns, inconsistent design system usage
+- **Info**: polish/consistency opportunities
+
+Score each category 1-10 and include evidence (`file:line` for code audits).
+
+## Phase 4: Output
+
+Return:
+
+1. **Result**: PASS / NEEDS WORK
+2. **Audited files** (list)
+3. **Category scores**
+4. **Findings by severity** with actionable fixes
+5. **Fast remediation plan** (top 3 fixes first)
+
+If `--full-report` is false, omit empty categories.
+
+## Record Findings
+
+```typescript
+observation({
+  type: "warning",
+  title: "UI Slop Check: [scope]",
+  narrative: "Detected [count] critical, [count] warning slop issues in changed UI files.",
+  concepts: "ui, design, anti-patterns, frontend",
+  confidence: "high",
+});
+```
+
+## Related Commands
+
+| Need                                     | Command      |
+| ---------------------------------------- | ------------ |
+| Design from scratch                      | `/design`    |
+| Full UI review (single screen/component) | `/ui-review` |
+| Implementation work                      | `/ship`      |

package/dist/template/.opencode/command/verify.md

@@ -0,0 +1,173 @@
+---
+description: Verify implementation completeness, correctness, and coherence
+argument-hint: "<bead-id> [--quick] [--full] [--fix] [--no-cache]"
+agent: review
+---
+
+# Verify: $ARGUMENTS
+
+Check implementation against PRD before shipping.
+
+## Load Skills
+
+```typescript
+skill({ name: "beads" });
+skill({ name: "verification-before-completion" });
+// If local web/browser verification needs stable URLs: skill({ name: "portless" });
+```
+
+## Parse Arguments
+
+| Argument     | Default  | Description                                    |
+| ------------ | -------- | ---------------------------------------------- |
+| `<bead-id>`  | required | The bead to verify                             |
+| `--quick`    | false    | Gates only, skip coherence check               |
+| `--full`     | false    | Force full verification mode (non-incremental) |
+| `--fix`      | false    | Auto-fix lint/format issues                    |
+| `--no-cache` | false    | Bypass verification cache, force fresh run     |
+
+## Determine Input Type
+
+| Input Type | Detection                   | Action                              |
+| ---------- | --------------------------- | ----------------------------------- |
+| Bead ID    | Matches `br-xxx` or numeric | Check implementation vs PRD in bead |
+| Path       | File/directory path         | Verify that specific path           |
+| `all`      | Keyword                     | Verify all in-progress work         |
+
+## Before You Verify
+
+- **Be certain**: Only flag issues you can verify with tools
+- **Don't invent problems**: If an edge case isn't in the PRD, don't flag it
+- **Run the gates**: Build, test, lint, typecheck are non-negotiable
+- **Use project conventions**: Check `package.json` scripts first
+
+## Phase 0: Check Verification Cache
+
+Before running any gates, check if a recent verification is still valid:
+
+```bash
+# Compute current state fingerprint (commit hash + diff)
+CURRENT_STAMP=$(printf '%s\n%s' \
+  "$(git rev-parse HEAD)" \
+  "$(git diff HEAD -- '*.ts' '*.tsx' '*.js' '*.jsx')" \
+  | shasum -a 256 | cut -d' ' -f1)
+LAST_STAMP=$(tail -1 .beads/verify.log 2>/dev/null | awk '{print $1}')
+```
+
+| Condition                                 | Action                                                 |
+| ----------------------------------------- | ------------------------------------------------------ |
+| `--no-cache` or `--full`                  | Skip cache check, run fresh                            |
+| `CURRENT_STAMP == LAST_STAMP`             | Report **cached PASS**, skip to Phase 2 (completeness) |
+| `CURRENT_STAMP != LAST_STAMP` or no cache | Run gates normally                                     |
+
+When cache hits, report:
+
+```text
+Verification: cached PASS (no changes since <timestamp from verify.log>)
+```
+
+## Phase 1: Gather Context
+
+```bash
+br show $ARGUMENTS
+```
+
+Read `.beads/artifacts/$ARGUMENTS/` to check what artifacts exist.
+
+Read the PRD and any other artifacts (plan.md, research.md, design.md).
+
+**Verify guards:**
+
+- [ ] Bead is `in_progress`
+- [ ] `prd.md` exists
+- [ ] You have read the full PRD
+
+## Phase 2: Completeness
+
+Extract all requirements/tasks from the PRD and verify each is implemented:
+
+- For each requirement: find evidence in the codebase (file:line reference)
+- Mark as: complete, partial, or missing
+- Report completeness score (X/Y requirements met)
+
+## Phase 3: Correctness
+
+Follow the [Verification Protocol](../skill/verification-before-completion/references/VERIFICATION_PROTOCOL.md):
+
+**Default: incremental mode** (changed files only, parallel gates).
+
+| Mode        | When                                      | Behavior                         |
+| ----------- | ----------------------------------------- | -------------------------------- |
+| Incremental | Default, <20 changed files                | Lint changed files, test changed |
+| Full        | `--full` flag, >20 changed files, or ship | Lint all, test all               |
+
+**Execution order:**
+
+1. **Parallel**: typecheck + lint (simultaneously)
+2. **Sequential** (after parallel passes): test, then build (ship only)
+
+For browser/manual local-web requirements, load the [portless](../skill/portless/SKILL.md) skill when stable named URLs would improve verification. Use only read-only Portless commands unless the user explicitly approves local networking changes; a reachable Portless URL supplements, but never replaces, typecheck/lint/test/build evidence.
+
+Report results with mode column:
+
+```text
+| Gate      | Status | Mode        | Time   |
+|-----------|--------|-------------|--------|
+| Typecheck | PASS   | full        | 2.1s   |
+| Lint      | PASS   | incremental | 0.3s   |
+| Test      | PASS   | incremental | 1.2s   |
+| Build     | SKIP   | —           | —      |
+```
+
+**After all gates pass**, record to verification cache:
+
+```bash
+echo "$CURRENT_STAMP $(date -u +%Y-%m-%dT%H:%M:%SZ) PASS" >> .beads/verify.log
+```
+
+If `--fix` flag provided, run the project's auto-fix command (e.g., `npm run lint:fix`, `ruff check --fix`, `cargo clippy --fix`).
+
+## Phase 4: Coherence (skip with --quick)
+
+Cross-reference artifacts for contradictions:
+
+- PRD vs implementation (does code address all PRD requirements?)
+- Plan vs implementation (did code follow the plan?)
+- Research recommendations vs actual approach (if different, is it justified?)
+
+Flag contradictions with specific file references.
+
+## Phase 5: Report
+
+```bash
+br comments add $ARGUMENTS "Verification: [PASS|PARTIAL|FAIL] - [summary]"
+```
+
+Output:
+
+1. **Result**: READY TO SHIP / NEEDS WORK / BLOCKED
+2. **Completeness**: score and status
+3. **Correctness**: gate results (with mode column)
+4. **Coherence**: contradictions found (if not --quick)
+5. **Blocking issues** to fix before shipping
+6. **Next step**: `/ship $ARGUMENTS` if ready, or list fixes needed
+
+Record significant findings with `observation()`:
+
+```typescript
+observation({
+  type: "discovery", // or "warning", "bugfix"
+  title: "Verify: [bead-id] [key finding]",
+  narrative: "[What was found, impact, resolution]",
+  concepts: "verification, [component]",
+  confidence: "high",
+});
+```
+
+## Related Commands
+
+| Need              | Command            |
+| ----------------- | ------------------ |
+| Ship after verify | `/ship <id>`       |
+| Review code       | `/review-codebase` |
+| Check status      | `/status`          |

package/dist/template/.opencode/context/README.md

@@ -0,0 +1,29 @@
+# Project Context
+
+Files in this directory are automatically injected into all AI prompts via the `instructions` array in `opencode.json`.
+
+## Purpose
+
+Put project-specific context here that you want the AI to always know about:
+- Architecture decisions
+- Business domain knowledge  
+- API conventions
+- Team agreements
+- Anything the AI needs as background context
+
+## How it works
+
+1. Add `.md` files to this directory
+2. Add the file path to `opencode.json` → `instructions[]`:
+   ```json
+   "instructions": [
+     ".opencode/context/your-file.md"
+   ]
+   ```
+3. The file content will be injected into every AI prompt
+
+## Important
+
+- This directory is **preserved** during `init --force` and `upgrade`
+- Keep files concise — every token counts in the AI context window
+- Use markdown format for best results

package/dist/template/.opencode/context/git-context.md

@@ -0,0 +1,32 @@
+---
+purpose: Git spatial awareness injected into all agent prompts
+updated: 2026-04-01
+---
+
+# Git Context
+
+## Auto-Context (start of every task)
+
+Before starting work, establish your position in the repo. Run once per session and cache results:
+
+```bash
+git branch --show-current        # What branch am I on?
+git status --short               # What's modified/staged? (cap at 20 lines)
+git log --oneline -5             # What happened recently?
+```
+
+## When to Refresh
+
+Re-run git context after:
+
+- Any commit you make
+- Switching branches
+- Pulling, merging, or rebasing
+- Stashing or unstashing
+
+## How to Use
+
+- **Commit messages**: Use branch name for scope (e.g., `feat(auth):` if on `feature/auth`)
+- **Branch decisions**: Check if you're on main before creating a feature branch
+- **Conflict awareness**: `git status` shows modified files — avoid editing files with uncommitted changes from another task
+- **Recent history**: `git log` shows what was just done — prevents duplicate work

package/dist/template/.opencode/dcp-prompts/defaults/README.md

@@ -0,0 +1,40 @@
+# DCP Prompt Defaults
+
+This directory stores the DCP prompts.
+Each prompt file here should contain plain text only (no XML wrappers).
+
+## Creating Overrides
+
+1. Copy a prompt file from this directory into an overrides directory using the same filename.
+2. Edit the copied file using plain text.
+3. Restart OpenCode.
+
+To reset an override, delete the matching file from your overrides directory.
+
+Do not edit the default prompt files directly, they are just for reference, only files in the overrides directory are used.
+
+Override precedence (highest first):
+1. `.opencode/dcp-prompts/overrides/` (project)
+2. `$OPENCODE_CONFIG_DIR/dcp-prompts/overrides/` (config dir)
+3. `~/.config/opencode/dcp-prompts/overrides/` (global)
+
+## Prompt Files
+
+- `system.md`
+  - Purpose: Core system-level DCP instruction block.
+  - Runtime use: Injected into the model system prompt on every request.
+- `compress-range.md`
+  - Purpose: range-mode compress tool instructions and summary constraints.
+  - Runtime use: Registered as the range-mode compress tool description.
+- `compress-message.md`
+  - Purpose: message-mode compress tool instructions and summary constraints.
+  - Runtime use: Registered as the message-mode compress tool description.
+- `context-limit-nudge.md`
+  - Purpose: High-priority nudge when context is over max threshold.
+  - Runtime use: Injected when context usage is beyond configured max limits.
+- `turn-nudge.md`
+  - Purpose: Nudge to compress closed ranges at turn boundaries.
+  - Runtime use: Injected when context is between min and max limits at a new user turn.
+- `iteration-nudge.md`
+  - Purpose: Nudge after many iterations without user input.
+  - Runtime use: Injected when iteration threshold is crossed.

package/dist/template/.opencode/dcp-prompts/defaults/compress-message.md

@@ -0,0 +1,42 @@
+Collapse selected individual messages in the conversation into detailed summaries.
+
+THE SUMMARY
+Your summary must be EXHAUSTIVE. Capture file paths, function signatures, decisions made, constraints discovered, key findings, tool outcomes, and user intent details that matter... EVERYTHING that preserves the value of the selected message after the raw message is removed.
+
+USER INTENT FIDELITY
+When a selected message contains user intent, preserve that intent with extra care. Do not change scope, constraints, priorities, acceptance criteria, or requested outcomes.
+Directly quote short user instructions when that best preserves exact meaning.
+
+Yet be LEAN. Strip away the noise: failed attempts that led nowhere, verbose tool output, and repetition. What remains should be pure signal - golden nuggets of detail that preserve full understanding with zero ambiguity.
+If a message contains no significant technical decisions, code changes, or user requirements, produce a minimal one-line summary rather than a detailed one.
+
+MESSAGE IDS
+You specify individual raw messages by ID using the injected IDs visible in the conversation:
+
+- `mNNNN` IDs identify raw messages
+
+Each message has an ID inside XML metadata tags like `<dcp-message-id priority="high">m0007</dcp-message-id>`.
+The same ID tag appears in every tool output of the message it belongs to — each unique ID identifies one complete message.
+Treat these tags as message metadata only, not as content to summarize. Use only the inner `mNNNN` value as the `messageId`.
+The `priority` attribute indicates relative context cost. You MUST compress high-priority messages when their full text is no longer necessary for the active task.
+If prior compress-tool results are present, always compress and summarize them minimally only as part of a broader compression pass. Do not invoke the compress tool solely to re-compress an earlier compression result.
+Messages marked as `<dcp-message-id>BLOCKED</dcp-message-id>` cannot be compressed.
+
+Rules:
+
+- Pick each `messageId` directly from injected IDs visible in context.
+- Only use raw message IDs of the form `mNNNN`.
+- Ignore XML attributes such as `priority` when copying the ID; use only the inner `mNNNN` value.
+- Do not invent IDs. Use only IDs that are present in context.
+
+BATCHING
+Select MANY messages in a single tool call when they are safe to compress.
+Each entry should summarize exactly one message, and the tool can receive as many entries as needed in one batch.
+
+GENERAL CLEANUP
+Use the topic "general cleanup" for broad cleanup passes.
+During general cleanup, compress all medium and high-priority messages that are not relevant to the active task.
+Optimize for reducing context footprint, not for grouping messages by topic.
+Do not compress away still-active instructions, unresolved questions, or constraints that are likely to matter soon.
+Prioritize the earliest messages in the context as they will be the least relevant to the active task.
+General cleanup should be done periodically between other normal compression tool passes, not as the primary form of compression.

package/dist/template/.opencode/dcp-prompts/defaults/compress-range.md

@@ -0,0 +1,59 @@
+Collapse a range in the conversation into a detailed summary.
+
+THE SUMMARY
+Your summary must be EXHAUSTIVE. Capture file paths, function signatures, decisions made, constraints discovered, key findings... EVERYTHING that maintains context integrity. This is not a brief note - it is an authoritative record so faithful that the original conversation adds no value.
+
+USER INTENT FIDELITY
+When the compressed range includes user messages, preserve the user's intent with extra care. Do not change scope, constraints, priorities, acceptance criteria, or requested outcomes.
+Directly quote user messages when they are short enough to include safely. Direct quotes are preferred when they best preserve exact meaning.
+
+Yet be LEAN. Strip away the noise: failed attempts that led nowhere, verbose tool outputs, back-and-forth exploration. What remains should be pure signal - golden nuggets of detail that preserve full understanding with zero ambiguity.
+
+COMPRESSED BLOCK PLACEHOLDERS
+When the selected range includes previously compressed blocks, use this exact placeholder format when referencing one:
+
+- `(bN)`
+
+Compressed block sections in context are clearly marked with a header:
+
+- `[Compressed conversation section]`
+
+Compressed block IDs always use the `bN` form (never `mNNNN`) and are represented in the same XML metadata tag format.
+
+Rules:
+
+- Include every required block placeholder exactly once.
+- Do not invent placeholders for blocks outside the selected range.
+- Treat `(bN)` placeholders as RESERVED TOKENS. Do not emit `(bN)` text anywhere except intentional placeholders.
+- If you need to mention a block in prose, use plain text like `compressed bN` (not as a placeholder).
+- Preflight check before finalizing: the set of `(bN)` placeholders in your summary must exactly match the required set, with no duplicates.
+
+These placeholders are semantic references. They will be replaced with the full stored compressed block content when the tool processes your output.
+
+FLOW PRESERVATION WITH PLACEHOLDERS
+When you use compressed block placeholders, write the surrounding summary text so it still reads correctly AFTER placeholder expansion.
+
+- Treat each placeholder as a stand-in for a full conversation segment, not as a short label.
+- Ensure transitions before and after each placeholder preserve chronology and causality.
+- Do not write text that depends on the placeholder staying literal (for example, "as noted in `(b2)`").
+- Your final meaning must be coherent once each placeholder is replaced with its full compressed block content.
+
+BOUNDARY IDS
+You specify boundaries by ID using the injected IDs visible in the conversation:
+
+- `mNNNN` IDs identify raw messages
+- `bN` IDs identify previously compressed blocks
+
+Each message has an ID inside XML metadata tags like `<dcp-message-id>...</dcp-message-id>`.
+The same ID tag appears in every tool output of the message it belongs to — each unique ID identifies one complete message.
+Treat these tags as boundary metadata only, not as tool result content.
+
+Rules:
+
+- Pick `startId` and `endId` directly from injected IDs in context.
+- IDs must exist in the current visible context.
+- `startId` must appear before `endId`.
+- Do not invent IDs. Use only IDs that are present in context.
+
+BATCHING
+When multiple independent ranges are ready and their boundaries do not overlap, include all of them as separate entries in the `content` array of a single tool call. Each entry should have its own `startId`, `endId`, and `summary`.

package/dist/template/.opencode/dcp-prompts/defaults/context-limit-nudge.md

@@ -0,0 +1,15 @@
+CRITICAL WARNING: MAX CONTEXT LIMIT REACHED
+
+You are at or beyond the configured max context threshold. This is an emergency context-recovery moment.
+
+You MUST use the `compress` tool now. Do not continue normal exploration until compression is handled.
+
+If you are in the middle of a critical atomic operation, finish that atomic step first, then compress immediately.
+
+SELECTION PROCESS
+Start from older, resolved history and capture as much stale context as safely possible in one pass.
+Avoid the newest active working messages unless it is clearly closed.
+
+SUMMARY REQUIREMENTS
+Your summary MUST cover all essential details from the selected messages so work can continue.
+If the compressed range includes user messages, preserve user intent exactly. Prefer direct quotes for short user messages to avoid semantic drift.

package/dist/template/.opencode/dcp-prompts/defaults/iteration-nudge.md

@@ -0,0 +1,3 @@
+You've been iterating for a while after the last user message.
+
+If there is a closed portion that is unlikely to be referenced immediately (for example, finished research before implementation), use the compress tool on it now.

package/dist/template/.opencode/dcp-prompts/defaults/system.md

@@ -0,0 +1,31 @@
+You operate in a context-constrained environment. Manage context continuously to avoid buildup and preserve retrieval quality. Efficient context management is paramount for your agentic performance.
+
+The ONLY tool you have for context management is `compress`. It replaces older conversation content with technical summaries you produce.
+
+`<dcp-message-id>` and `<dcp-system-reminder>` tags are environment-injected metadata. Do not output them.
+
+THE PHILOSOPHY OF COMPRESS
+`compress` transforms conversation content into dense, high-fidelity summaries. This is not cleanup - it is crystallization. Your summary becomes the authoritative record of what transpired.
+
+Think of compression as phase transitions: raw exploration becomes refined understanding. The original context served its purpose; your summary now carries that understanding forward.
+
+COMPRESS WHEN
+
+A section is genuinely closed and the raw conversation has served its purpose:
+
+- Research concluded and findings are clear
+- Implementation finished and verified
+- Exploration exhausted and patterns understood
+- Dead-end noise can be discarded without waiting for a whole chapter to close
+
+DO NOT COMPRESS IF
+
+- Raw context is still relevant and needed for edits or precise references
+- The target content is still actively in progress
+- You may need exact code, error messages, or file contents in the immediate next steps
+
+Before compressing, ask: _"Is this section closed enough to become summary-only right now?"_
+
+Evaluate conversation signal-to-noise REGULARLY. Use `compress` deliberately with quality-first summaries. Prioritize stale content intelligently to maintain a high-signal context window that supports your agency.
+
+It is of your responsibility to keep a sharp, high-quality context window for optimal performance.

package/dist/template/.opencode/dcp-prompts/defaults/turn-nudge.md

@@ -0,0 +1,7 @@
+Evaluate the conversation for compressible ranges.
+
+If any messages are cleanly closed and unlikely to be needed again, use the compress tool on them.
+If direction has shifted, compress earlier ranges that are now less relevant.
+
+The goal is to filter noise and distill key information so context accumulation stays under control.
+Keep active context uncompressed.