opencodekit 0.19.5 → 0.20.0

package/dist/template/.opencode/skill/writing-skills/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: writing-skills
-description: "Use when creating new skills, editing existing skills, or verifying skills work before deployment - applies TDD to process documentation by testing with subagents before writing, iterating until bulletproof against rationalization"
+description: "Use when creating new skills, editing existing skills, or verifying skills work before deployment - applies TDD to process documentation by testing with subagents before writing, iterating until bulletproof against rationalization. Includes complete pressure testing methodology."
 version: 1.0.0
 tags: [documentation, workflow]
 dependencies: []
@@ -211,12 +211,48 @@ Run same scenarios WITH skill. Agent should now comply.
 
 Agent found new rationalization? Add explicit counter. Re-test until bulletproof.
 
-**REQUIRED SUB-SKILL:** Use skill({ name: "testing-skills-with-subagents" }) for the complete testing methodology:
+**DETAILED TESTING METHODOLOGY:** See `references/testing-methodology.md` for the complete testing guide:
 
-- How to write pressure scenarios
-- Pressure types (time, sunk cost, authority, exhaustion)
-- Plugging holes systematically
-- Meta-testing techniques
+- Full RED-GREEN-REFACTOR testing workflow for skills
+- Scenario templates and pressure campaign examples
+- Rationalization hardening patterns and iteration loops
+- End-to-end checklists and worked examples
+
+### Inline Testing Summary (Merged Essentials)
+
+Use this quick inline summary while authoring; use `references/testing-methodology.md` for full details.
+
+#### Pressure scenarios (test format)
+
+- Test behavior under realistic stress, not quiz-style prompts
+- Force concrete A/B/C choices so violations are observable
+- Combine at least 3 pressures per scenario for discipline skills
+- Capture exact rationalizations verbatim during RED
+
+| Pressure Type   | What to Inject                              |
+| --------------- | ------------------------------------------- |
+| Time            | Deadline, production incident, deploy clock |
+| Sunk cost       | Hours already spent, large diff to discard  |
+| Authority       | Senior/manager asking to skip process       |
+| Economic        | Revenue loss, promotion/job pressure        |
+| Exhaustion      | End-of-day fatigue, urgency to finish       |
+| Social          | Fear of looking rigid or uncooperative      |
+| Pragmatic frame | "Be practical, not dogmatic" framing        |
+
+#### Meta-testing (after failure)
+
+When agent fails with skill loaded, ask how the skill should be rewritten to make the right choice unambiguous.
+
+- If response is "skill was clear, I ignored it" → strengthen foundational principle
+- If response is "skill should explicitly say X" → add X verbatim
+- If response is "I missed section Y" → improve placement/prominence
+
+#### Exit criteria for bulletproof skills
+
+- Agent chooses correct action under maximum pressure
+- Agent cites relevant sections to justify action
+- No new rationalizations emerge across retests
+- Meta-test confirms clarity, not ambiguity
 
 ## STOP: Before Moving to Next Skill
 
@@ -281,6 +317,7 @@ Deploying untested skills = deploying untested code. It's a violation of quality
 - `references/claude-search-optimization.md` - CSO guidance: descriptions, keywords, token efficiency, cross-references
 - `references/flowcharts-and-examples.md` - Flowchart usage rules and code example guidance
 - `references/file-organization.md` - Patterns for self-contained vs heavy-reference skills
+- `references/testing-methodology.md` - Complete pressure-testing methodology merged from testing-skills-with-subagents
 - `references/testing-skill-types.md` - How to test discipline, technique, pattern, and reference skills
 - `references/rationalization-hardening.md` - Loophole closure, rationalization tables, red flags
 - `references/anti-patterns.md` - Anti-patterns to avoid

package/dist/template/.opencode/skill/{testing-skills-with-subagents/SKILL.md → writing-skills/references/testing-methodology.md}

@@ -1,11 +1,3 @@
----
-name: testing-skills-with-subagents
-description: Use when creating or editing skills, before deployment, to verify they work under pressure and resist rationalization - applies RED-GREEN-REFACTOR cycle to process documentation by running baseline without skill, writing to address failures, iterating to close loopholes
-version: 1.0.0
-tags: [testing, agent-coordination]
-dependencies: [writing-skills]
----
-
 # Testing Skills With Subagents
 
 ## When to Use

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "opencodekit",
-	"version": "0.19.5",
+	"version": "0.20.0",
 	"description": "CLI tool for bootstrapping and managing OpenCodeKit projects",
 	"keywords": [
 		"agents",

package/dist/template/.opencode/skill/compaction/SKILL.md

@@ -1,384 +0,0 @@
----
-name: compaction
-description: >
-  Use when context is growing large during long-running tasks and needs server-side or client-side
-  summarization to continue effectively. Covers compaction triggers, custom summarization patterns,
-  session handoff, and context preservation strategies.
-version: 1.0.0
-tags: [context, workflow]
-dependencies: []
----
-
-# Context Compaction - Managing Long-Running Sessions
-
-## When to Use
-
-- Context is growing large and needs compression/cleanup to continue.
-
-## When NOT to Use
-
-- Short sessions with low context usage where no compaction is needed.
-
-## Overview
-
-**Compaction = Summarization + Preservation + Continuity**
-
-Long-running sessions accumulate context (tool outputs, code reads, exploration results). When context approaches limits, compaction reduces it to essential information while preserving decision history and work state.
-
-## Context Budget Awareness
-
-### Token Thresholds
-
-| Context Usage | Status      | Action                                        |
-| ------------- | ----------- | --------------------------------------------- |
-| 0-50%         | 🟢 Normal   | Work freely                                   |
-| 50-70%        | 🟡 Watch    | Start compressing completed explorations      |
-| 70-85%        | 🟠 Compact  | Actively compress and sweep stale noise       |
-| 85-95%        | 🔴 Critical | Emergency compaction, prepare session handoff |
-| 95%+          | ⛔ Limit    | Session handoff required                      |
-
-### Monitoring
-
-Pay attention to these signals:
-
-- Completed phases accumulating without being compressed
-- Repeated file reads of the same content
-- Large bash outputs from builds/tests
-- Multiple exploration rounds without synthesis
-
-## Compaction Strategies
-
-### Strategy 1: Phase Compression (Preferred)
-
-Compress completed conversation phases into dense summaries. This is the primary DCP instrument in the installed beta.
-
-Compress completed conversation phases into dense summaries.
-
-```
-WHEN: A research phase is complete and findings are clear
-DO: compress({
-  topic: "Auth Research Complete",
-  content: {
-    startString: "unique text at phase start",
-    endString: "unique text at phase end",
-    summary: "Complete technical summary of findings..."
-  }
-})
-```
-
-**Key principle**: Only compress CLOSED chapters. Never compress active work.
-
-### Strategy 2: Sweep Stale Noise
-
-Use `/dcp sweep` after a phase is complete to remove stale/noisy content automatically.
-
-```
-WHEN: Wrong-target searches or superseded outputs are no longer needed
-DO: /dcp sweep
-```
-
-**Key principle**: Sweep only after the relevant phase is closed. If in doubt, keep it.
-
-### Strategy 3: Session Handoff
-
-When context is too large to compact further, hand off to a new session.
-
-```
-WHEN: Context > 85% and significant work remains
-DO:
-  1. Create handoff document with memory-update
-  2. Save all decisions with observation tool
-  3. Document current state and remaining work
-  4. Start new session with handoff reference
-```
-
-## Compaction Decision Tree
-
-```
-Is context growing large?
-├── NO → Continue working normally
-└── YES → What type of content is consuming space?
-    ├── Completed conversation phases → COMPRESS
-    ├── Stale/noisy closed-phase outputs → SWEEP
-    └── Everything is still relevant → SESSION HANDOFF
-```
-
-## Custom Summarization Patterns
-
-### For Code Exploration
-
-```markdown
-## Exploration Summary: [Component/Module]
-
-### Architecture
-
-- Entry point: `src/auth/index.ts`
-- Key classes: AuthService, TokenManager, SessionStore
-- Dependencies: jwt, bcrypt, redis
-
-### Key Findings
-
-- Auth flow: login → validate → issue JWT → store session
-- Token rotation: every 15 minutes via refresh endpoint
-- Session storage: Redis with 24h TTL
-
-### Decisions Made
-
-- Use existing TokenManager (don't replace)
-- Add rate limiting to login endpoint
-- Migrate session store from memory to Redis
-
-### Files to Modify
-
-- src/auth/service.ts (add rate limiting)
-- src/auth/session.ts (Redis integration)
-- src/config/redis.ts (new file)
-```
-
-### For Implementation Phase
-
-```markdown
-## Implementation Summary: [Feature]
-
-### Completed
-
-- [x] Database schema migration (src/db/migrations/004_auth.ts)
-- [x] API endpoints (src/routes/auth.ts) - 3 new routes
-- [x] Frontend forms (src/components/auth/) - Login, Register, Reset
-
-### Verification
-
-- TypeScript: ✅ passing
-- Tests: ✅ 12/12 passing
-- Lint: ✅ no issues
-
-### Remaining
-
-- [ ] Email verification flow
-- [ ] Rate limiting middleware
-
-### Key Decisions
-
-- JWT expiry: 15 minutes (refresh: 7 days)
-- Password hashing: bcrypt with 12 rounds
-- Session storage: Redis (not in-memory)
-```
-
-### For Debugging
-
-```markdown
-## Debug Summary: [Issue]
-
-### Symptoms
-
-- Error: "TypeError: Cannot read property 'id' of undefined"
-- Location: src/auth/middleware.ts:42
-- Trigger: POST /api/protected when token is expired
-
-### Root Cause
-
-- Token validation returns null on expired tokens
-- Middleware assumes valid token object, no null check
-- Race condition: token expires between validation and use
-
-### Fix Applied
-
-- Added null check in middleware (src/auth/middleware.ts:42)
-- Added token refresh attempt before rejecting (src/auth/refresh.ts)
-- Added test for expired token scenario (src/auth/**tests**/middleware.test.ts)
-
-### Verification
-
-- Tests: ✅ all passing including new test
-- Manual: ✅ expired token now triggers refresh
-```
-
-## Session Handoff Protocol
-
-When you must hand off to a new session:
-
-### 1. Create Handoff Document
-
-```typescript
-memory -
-  update({
-    file: "handoffs/YYYY-MM-DD-feature-name",
-    content: `# Session Handoff: [Feature Name]
-
-## Context
-[Why this session started, what was the goal]
-
-## Completed Work
-[What was done, files changed, decisions made]
-
-## Current State
-[Where things stand right now]
-
-## Remaining Work
-[What still needs to be done]
-
-## Key Decisions
-[Important choices made and why]
-
-## Files Modified
-[List of all files changed with brief description]
-
-## Gotchas
-[Things the next session should know]
-  `,
-    mode: "replace",
-  });
-```
-
-### 2. Save Key Observations
-
-```typescript
-observation({
-  type: "decision",
-  title: "Auth implementation approach",
-  narrative: "Chose JWT with Redis sessions because...",
-  facts: "JWT 15min expiry, Redis 24h TTL, bcrypt 12 rounds",
-  concepts: "authentication, sessions, tokens",
-  confidence: "high",
-});
-```
-
-### 3. Resume in New Session
-
-```typescript
-// In new session:
-memory - read({ file: "handoffs/YYYY-MM-DD-feature-name" });
-memory - search({ query: "auth implementation" });
-```
-
-## Integration with DCP Plugin
-
-This project uses `@tarquinen/opencode-dcp` for always-on context management (injected via `experimental.chat.system.transform`):
-
-- **compress**: Phase-level conversation compression (primary DCP tool in the installed beta)
-- **/dcp sweep**: Cleanup command for stale/noisy content after a phase is complete
-- **Prunable-tools list**: Auto-injected into messages with token estimates
-- **Nudge system**: Reminders every N tool calls + critical limit warnings
-
-**Division of responsibility:**
-
-- **DCP plugin**: Context budget rules, tool guidance, prunable-tools list, nudges (always present via system prompt)
-- **Compaction plugin** (`.opencode/plugin/compaction.ts`): Session continuity, beads state, handoff recovery, post-compaction protocol (fires during compaction events only)
-
-**Custom prompts (enabled):**
-
-DCP custom prompts are enabled (`experimental.customPrompts: true`). Override files live at `.opencode/dcp-prompts/overrides/`:
-
-- `compress-message.md` — 9-section structured summary format (Primary Request, Key Technical Concepts, Files/Code, Errors/Fixes, Problem Solving, User Messages, Pending Tasks, Current Work, Next Step)
-
-Override precedence: project (`.opencode/dcp-prompts/overrides/`) > config dir > global (`~/.config/opencode/dcp-prompts/overrides/`).
-
-## Post-Compaction Restoration Protocol
-
-**Critical:** After any compaction (server-side or DCP compress), you lose raw file content and active context. Immediately restore:
-
-### Step 1: Re-read Active Files (max 5)
-
-Re-read the files you were actively editing before compaction. Prioritize by recency — most recently modified first. Cap at 5 files, use offset/limit to read only the sections you need (~200 lines each).
-
-```
-After compaction, if you were editing src/auth.ts:
-read({ filePath: "src/auth.ts", offset: 30, limit: 50 })
-```
-
-**Why:** The Write/Edit tool requires a prior Read in the same session (`FileTime.assert()`). Without re-reading, your next edit attempt will fail.
-
-### Step 2: Restore Active Skill Context
-
-If you were using a skill before compaction, re-load it:
-
-```typescript
-skill({ name: "<active-skill>" });
-```
-
-Skills are prompt-injected — compaction drops them. Re-load to restore the workflow.
-
-### Step 3: Check Todo State
-
-Review your TodoWrite list to re-establish task tracking:
-
-```
-- What was in_progress?
-- What was pending?
-- What was completed?
-```
-
-### Step 4: Verify Git Position
-
-```bash
-git branch --show-current
-git status --short
-```
-
-Confirms you're on the right branch and shows any uncommitted changes.
-
-### Step 5: Scan Memory for Context
-
-If the compaction was heavy (>50% reduction), check memory for recent observations:
-
-```typescript
-memory - search({ query: "<current task keywords>", limit: 3 });
-```
-
-### Why This Protocol Exists
-
-Compaction strips raw file content, tool outputs, and skill prompts. Without restoration:
-
-- File edits fail (`FileTime.assert` — no prior read)
-- Line-number precision is lost (you're guessing, not citing)
-- Skill workflows are inactive (no prompt injection)
-- Task tracking is disconnected (todos lost in compressed summary)
-
-## Anti-Patterns
-
-### ❌ Premature Compaction
-
-```
-// DON'T compress a file you're about to edit
-compress({ ... })  // Loses exact line numbers you need
-edit({ ... })      // Now you can't find the right location
-```
-
-**Fix**: Keep raw content while actively editing. Compress AFTER the edit phase.
-
-### ❌ Compressing Active Work
-
-```
-// DON'T compress a conversation phase you might return to
-compress({ summary: "Explored auth options" })
-// Later: "Wait, which options did we consider?"
-```
-
-**Fix**: Only compress CLOSED chapters where findings are crystallized.
-
-### ❌ Ignoring Context Growth
-
-```
-// DON'T let context grow unchecked until hitting limits
-// By the time you notice, emergency compaction loses information
-```
-
-**Fix**: Monitor regularly. Compress at natural breakpoints and use `/dcp sweep` after closed phases.
-
-## Checklist
-
-Before compacting:
-
-- [ ] Identified what type of content is consuming context
-- [ ] Chosen appropriate strategy (compress/sweep/handoff)
-- [ ] Verified raw content is no longer needed for active work
-- [ ] Captured all key details in the compression summary or handoff
-- [ ] Saved important decisions as observations
-- [ ] Created handoff document if switching sessions
-
-During long sessions:
-
-- [ ] Compressing completed phases at natural breakpoints
-- [ ] Sweeping stale/noisy outputs after phase completion
-- [ ] Monitoring context usage trends
-- [ ] Planning session handoff if approaching limits

package/dist/template/.opencode/skill/session-management/SKILL.md

@@ -1,9 +0,0 @@
----
-description: "Merged into context-management. Load that skill instead."
----
-
-# Session Management
-
-> This skill has been merged into `context-management` which now covers both context and session lifecycle management.
-> 
-> Load `context-management` instead.