opencodekit 0.21.9 → 0.22.0

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

@@ -0,0 +1,66 @@
+---
+description: Write or generate tests for code
+argument-hint: "<file or function> [--unit|--integration|--e2e]"
+agent: build
+---
+
+# Test: $ARGUMENTS
+
+Write or generate tests for the specified code.
+
+## Load Skills
+
+```typescript
+skill({ name: "test-driven-development" });
+skill({ name: "testing-anti-patterns" });
+skill({ name: "verification-before-completion" });
+```
+
+## Parse Arguments
+
+| Argument | Default | Description |
+|---|---|---|
+| `<file or function>` | required | File path or function name to test |
+| `--unit` | auto | Unit test (default for utils, helpers) |
+| `--integration` | false | Integration test (default for API, DB) |
+| `--e2e` | false | End-to-end test (requires full setup) |
+
+## Process
+
+### Phase 1: Understand
+
+1. Read the file and understand the function/module contract
+2. Identify: inputs, outputs, side effects, error states, edge cases
+3. Check existing tests for patterns and structure
+
+### Phase 2: Write Tests
+
+Cover:
+- Happy path (expected input → expected output)
+- Edge cases (empty, null, boundary values)
+- Error states (invalid input, missing dependencies)
+- Contract invariants (what must always be true)
+
+Follow the existing test patterns in the project. If none exist, check `vitest.config.ts`.
+
+### Phase 3: Run and Verify
+
+```bash
+vitest <path>  # or npm test
+```
+
+TDD if writing implementation simultaneously: RED → GREEN → REFACTOR.
+
+## Anti-Patterns
+
+- Testing implementation details instead of behavior
+- Mocking without understanding the real dependency
+- Test-only code in production files (exporting internals just for tests)
+- Brittle assertions (matching full objects when partial match suffices)
+
+## Output
+
+Report:
+1. Tests written
+2. Coverage summary
+3. Verification results

package/dist/template/.opencode/dcp.jsonc

@@ -1,18 +1,22 @@
 {
 	"$schema": "https://raw.githubusercontent.com/Opencode-DCP/opencode-dynamic-context-pruning/master/dcp.schema.json",
 	"enabled": true,
+	// Disable auto-update for config stability — update explicitly when pinning new versions.
+	// v3.1.10+ auto-update is mature; set true if you want rolling bug fixes.
+	"autoUpdate": false,
 	"debug": false,
 	// "off" | "minimal" | "detailed" — keep minimal for low-noise dev flow
 	"pruneNotification": "minimal",
 	// "chat" (in-conversation) or "toast" (system notification)
 	"pruneNotificationType": "toast",
-	// Slash commands: /dcp context, /dcp stats, /dcp sweep, /dcp compress, /dcp decompress, /dcp recompress
+	// Slash commands: /dcp context, /dcp stats, /dcp sweep, /dcp compress,
+	//                 /dcp decompress, /dcp recompress, /dcp manual
 	"commands": {
 		"enabled": true,
 		// Additional tools to protect from /dcp sweep (supports glob wildcards)
 		"protectedTools": ["observation", "memory-*"]
 	},
-	// Manual mode: disables autonomous context management
+	// Manual mode: when enabled, tools only run via /dcp commands — no autonomous pruning
 	"manualMode": {
 		"enabled": false,
 		"automaticStrategies": true
@@ -22,17 +26,15 @@
 		"enabled": false,
 		"turns": 4
 	},
-	// Glob patterns for files that should never be auto-pruned
-	// Keep tight: broad patterns reduce DCP effectiveness
-	// .opencode/** and .beads/** removed — memory-* and tilth_* outputs
-	// already survive compression via compress.protectedTools
+	// Glob patterns matched against tool parameters.filePath — keep tight
+	// Broad patterns reduce DCP effectiveness
 	"protectedFilePatterns": [
 		"**/.env*",
 		"**/AGENTS.md",
+		"**/opencode.json*",
 		"**/package.json",
 		"**/tsconfig.json"
 	],
-	// Unified context compression tool (v3.1.0)
 	"compress": {
 		// "range" (stable) compresses spans into block summaries
 		// "message" (experimental) compresses individual raw messages
@@ -40,45 +42,14 @@
 		// "allow" (no prompt) | "ask" (prompt) | "deny" (tool not registered)
 		"permission": "allow",
 		"showCompression": false,
-		// v3.1.0: active summary tokens extend effective maxContextLimit
+		// v3.1.0+: active summary tokens extend effective maxContextLimit
 		"summaryBuffer": true,
+		// v3.1.10+: percentage strings now use the model input budget directly —
+		// safe for GitHub Copilot and all other providers (no modelContextLimit required)
 		// Soft upper threshold: above this, strong compression nudges fire
-		// Use numeric values — percentage requires modelContextLimit from runtime
-		// which may be unavailable for some provider/model combos (e.g. GitHub Copilot)
-		// Rule: must be BELOW OpenCode emergency threshold (model_max - reserved - max_output)
-		// For Copilot Claude (216k ctx, 64k out, 16k reserved): emergency = 136k
-		// So DCP must start compressing well before 136k
-		"maxContextLimit": 100000,
-		// Per-model override for maxContextLimit (takes priority over global)
-		"modelMaxLimits": {
-			// Claude: 216k ctx, 64k out → emergency at 136k → DCP starts at 110k
-			"github-copilot/claude-opus-4.6": 110000,
-			"github-copilot/claude-opus-4.5": 110000,
-			"github-copilot/claude-sonnet-4.6": 110000,
-			"github-copilot/claude-sonnet-4.5": 110000,
-			"github-copilot/claude-sonnet-4": 110000,
-			// Haiku: smaller model, be more conservative
-			"github-copilot/claude-haiku-4.5": 90000,
-			// GPT/Gemini: assume similar 200k+ windows
-			"github-copilot/gpt-5.4": 110000,
-			"github-copilot/gpt-5.3-codex": 110000,
-			"github-copilot/gemini-3.1-pro-preview": 110000
-		},
+		"maxContextLimit": "75%",
 		// Soft lower threshold: below this, turn/iteration reminders are off
-		// Use numeric values — same reason as maxContextLimit above
-		"minContextLimit": 50000,
-		// Per-model override for minContextLimit (takes priority over global)
-		"modelMinLimits": {
-			"github-copilot/claude-opus-4.6": 65000,
-			"github-copilot/claude-opus-4.5": 65000,
-			"github-copilot/claude-sonnet-4.6": 65000,
-			"github-copilot/claude-sonnet-4.5": 65000,
-			"github-copilot/claude-sonnet-4": 65000,
-			"github-copilot/claude-haiku-4.5": 50000,
-			"github-copilot/gpt-5.4": 65000,
-			"github-copilot/gpt-5.3-codex": 65000,
-			"github-copilot/gemini-3.1-pro-preview": 65000
-		},
+		"minContextLimit": "30%",
 		// How often context-limit nudge fires above maxContextLimit (1 = every fetch)
 		"nudgeFrequency": 5,
 		// Messages since last user message before adding compression reminders
@@ -87,11 +58,23 @@
 		"nudgeForce": "soft",
 		// Keep user messages compressible to avoid permanent context growth
 		"protectUserMessages": false,
-		// Auto-protected by DCP: task, skill, todowrite, todoread, compress, batch, plan_enter, plan_exit, write, edit
+		// v3.1.11+: preserve text wrapped in <protect>...</protect> during compression
+		"protectTags": false,
+		// Optional per-model max context limits — only if you need different thresholds
+		// per model. Uncomment and adjust for your models:
+		// "modelMaxLimits": {
+		//     "anthropic/claude-sonnet-4.6": "80%",
+		//     "openai/gpt-5.3-codex": "80%",
+		//     "github/github-copilot": "65%"
+		// },
+		// "modelMinLimits": {
+		//     "anthropic/claude-sonnet-4.6": "30%"
+		// },
+		// Auto-protected by DCP: task, skill, todowrite, todoread, compress, batch,
+		//                        plan_enter, plan_exit, write, edit
 		// Only list ADDITIONAL tools whose outputs should be appended to compression summaries
-		"protectedTools": ["observation", "memory-*", "tilth_*"]
+		"protectedTools": ["observation", "memory-*"]
 	},
-	// Experimental features
 	"experimental": {
 		// Allow DCP processing in subagent sessions (default: false)
 		"allowSubAgents": false,

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

@@ -38,11 +38,9 @@ memory/
 
 | Tool                | Purpose                                       |
 | ------------------- | --------------------------------------------- |
-| `observation()`     | Write structured observations to SQLite       |
-| `memory-search()`   | FTS5 full-text search (returns compact index) |
-| `memory-get()`      | Fetch full observation details by ID          |
-| `memory-timeline()` | Get chronological context around anchor       |
-| `memory-update()`   | Update markdown files in memory               |
+| `observation()`     | Write structured observations to SQLite or give feedback on existing ones |
+| `memory-search()`   | FTS5 full-text search (returns compact index) or read memory files by path |
+| `memory-admin()`    | Memory system administration: maintenance, migration, lint, compile, index |
 
 ## PARA Categories
 

package/dist/template/.opencode/memory/_templates/adr.md

@@ -0,0 +1,45 @@
+# ADR: [short title of decision]
+
+**Status:** [proposed | accepted | deprecated | superseded]
+**Date:** [YYYY-MM-DD]
+**Context:** [bead-id, issue/Linear ticket, or session reference]
+
+## Context
+
+What is the problem we are solving? What forces are at play? What constraints do we have?
+
+## Decision
+
+What did we decide to do? Be specific. This is the answer to the problem stated above.
+
+## Rationale
+
+Why this approach over the alternatives? What evidence or reasoning supports this decision?
+
+## Consequences
+
+What becomes easier or harder because of this decision? What tradeoffs did we accept?
+
+### Positive
+
+- [benefit 1]
+- [benefit 2]
+
+### Negative
+
+- [cost / risk 1]
+- [cost / risk 2]
+
+## Alternatives Considered
+
+### Alternative 1: [name]
+
+Pros/cons and why it was rejected.
+
+### Alternative 2: [name]
+
+Pros/cons and why it was rejected.
+
+## Notes
+
+- [Anything else? Links to related ADRs, PRDs, specs?]

package/dist/template/.opencode/memory/project/gotchas.md

@@ -53,7 +53,7 @@ The edit tool (`str_replace`) is the #1 source of failures in LLM coding. Models
 ## Memory System
 
 - Subagents (explore, scout, review) should NOT write to memory - only leader agents
-- Use `observation: false` and `memory-update: false` in agent configs to enforce
+- Use `observation: false` in agent configs to enforce
 
 ## Build System
 

package/dist/template/.opencode/memory/project/user.md

@@ -30,8 +30,7 @@ updated: 2025-01-06
 ## Editing Tool Preferences
 
 - **Primary**: `edit` tool (str_replace) and `patch` tool
-- **Secondary/Fallback**: `tilth_tilth_edit` (hash-anchored edits) — only when str_replace fails
-- **Reading/Search**: `tilth_tilth_read` and `tilth_tilth_search` are fine to use freely
+- **Reading/Search**: `srcwalk` CLI via bash — `srcwalk find <symbol>`, `srcwalk <file>`, `srcwalk map`
 
 ## Rules to Always Follow
 

package/dist/template/.opencode/memory/session-context.md

@@ -6,7 +6,7 @@ updated: 2025-01-08
 # Session Context
 
 <!--
-Agent-maintained via memory-update tool.
+Agent-maintained via observation tool.
 Update when: goal changes, key decision made, state shifts, uncertainty discovered.
 After compaction: read this file, ask 1-3 targeted questions if gaps exist.
 -->