vibeostheog 0.13.22

package/CHANGELOG.md

@@ -0,0 +1,562 @@
+## 0.13.21
+- fix: blackbox save TTL throttle, export parseJsonc, stress/pattern tests
+- fix: rebuild bundled src/index.js with warn_count and ledger fixes
+- fix: warn_count tracking, est_savings_usd, and ledger reconcile
+
+
+## 0.13.20
+- fix: update deepseek/deepseek-v4-flash pricing to /bin/zsh.000146
+
+
+## 0.13.19
+- feat: trinity mode budget|quality|speed|longrun|auto command
+
+
+## 0.13.18
+- fix: per-session active_slot and optimization_mode in blackbox-state
+- fix: footer model tag shows active slot model instead of hardcoded brain tier
+- refactor: esbuild bundle — single-file vibeeOS.js output
+- chore: v0.13.17
+
+
+## 0.13.16
+- fix: auto-mode tier switch in footer (bypass broken system.transform hook)
+
+
+## 0.13.15
+- feat: progressive benchmark v3 — known-ground-truth scenarios, git isolation, full KPIs
+- feat: tier_bias now enforces active_slot switch via syncControlSettings
+- feat: nightly experiment runner for quality vs budget A/B testing
+- feat: syncControlSettings — mode-driven auto-toggle of all enforcement settings
+- fix: pass loadOptimizationMode() to computeControlVector so auto-mode actually switches tier
+- fix: ledger reconciliation reads usd field, writes total_savings_usd
+- fix: lifetime savings display and auto-mode tier routing
+- fix: align DFLT_SEL and loadSelection defaults with auto-mode
+- fix: align savings tracking fields across all systems
+- fix: optimization_mode now returns effective mode, not input value
+- fix: add setShellDirectory, fix flow-enforcer import path, shell.env directory wiring
+- fix: move applyDecadence import to state.js, add setLastMutationEvent
+- fix: const assignment + missing imports in shell.env hook
+- fix: move .ts stripping to end of deploy (after all copies)
+- fix: remove duplicate exports, add setLastMutationEvent, strip .ts on deploy
+- fix: replace all direct assignments to imported state vars with setters
+- fix: deploy .ts files (they are fixed, no stripping needed)
+- fix: deploy script now excludes .ts files from deployed plugin dir
+- fix: move applyDecadence to state.js where decadence vars are owned
+- fix: missing exports, const assignment bug in applyDecadence, TDD test expectations
+- fix: missing TRINITY_CHEAP/MEDIUM imports in hook files — plugin crashed on load
+- refactor: replace toy scenarios with 6 hard multi-file/architectural scenarios
+- refactor: regime-driven autoSelectMode, remove savings_goal_usd
+- docs: update README, AGENTS, LIVE_DEBUG for regime-driven autoSelectMode
+- docs: add LIVE_BUG section, ML/smart-cache tests, TDD live prompts
+- chore: sync compiled JS with TS savings fixes
+- chore: remove experiment artifacts, ignore experiments/ dir
+- chore: sync compiled index.js with TS source
+- chore: v0.13.11
+- chore: v0.13.10
+- chore: v0.13.9
+- chore: v0.13.8
+- chore: v0.13.7
+0.13.14
+0.13.13
+
+
+## 0.13.11
+- fix: use setters for all imported state vars (ES module bindings are read-only)
+- fix: add setLastMutationEvent, remove duplicate exports in state.js
+- fix: deploy script strips .ts files to prevent runtime conflicts
+- refactor: LIVE_DEBUG 136/137 pass, neutral env clean
+
+## 0.13.10
+- fix: move applyDecadence to state.js where _lastDecadenceRun is owned
+- fix: add missing TRINITY_CHEAP/MEDIUM, trendDisplay, scoreStress, remoteCall exports
+- docs: add LIVE_DEBUG.md with 87 prompt tests
+
+## 0.13.9
+- fix: deploy script missing src/lib/ — plugin hooks failed silently, no footer rendered
+
+## 0.13.8
+- docs: add LIVE_DEBUG.md with 87 prompt tests across 18 categories
+
+## 0.13.7
+- chore: release bump
+
+## 0.13.6
+- feat: web dashboard sidecar with SolidJS SPA, SSE, and standalone server
+- docs: add rule 7 — don't lie about tool execution, ask once
+
+
+## 0.13.5
+- refactor: extract 16 modules (7207 lines) from src/index.ts into src/lib/
+- fix: ES module binding divergence — add setters for mutable state
+- fix: missing imports, directory→projectDirectory rename, loadCredit import
+- fix: TDD auto-generated artifact cleanup
+
+## 0.13.4
+- feat: blackbox dynamically controls thinking mode per sub-regime for cost savings
+- feat: complete remote API migration — dual-path scoreStress, patternsObserve/Record, TDD exports with local fallback + neutral env test
+- feat: complete remote API migration — dual-path scoreStress, patternsObserve/Record, TDD exports with local fallback
+- feat: blackbox ML enhancements — real features, loop prevention, pivot detection, outcome tracking, calibration
+- feat: v0.10.0 — 6 enhancement phases implemented
+- feat: WordPress integration - atomic seat+token creation
+- feat: Phase 2 - Integrate remote API client into plugin runtime
+- feat: Phase 1 - Remote API server for protected algorithms
+- feat: CodeX MCP server and dashboard sidebar plugin integration
+- feat: vibeOS TUI dashboard sidebar plugin
+- fix: add setToolDirectory export to tool-execute.ts, skip pre-commit hooks in release.mjs
+- fix: release.mjs — add missing closing brace for deploy else block
+- fix: stabilize refactored modules — ES module bindings, setters, missing imports
+- fix: flow-enforcer race condition, blackbox default ON, dynamic footer
+- fix: lock model name, enforcement logging, TDD framework detection, cache display rounding
+- fix: validateState sessions object, remove stale report writes, drop dead code
+- fix: state validation, flow TODO dedup, session checkpointing, fetch verification
+- fix: _appendFooter full model names, → arrow, inline stress; 361/362 pass
+- fix: atomic state writes, safeJsonParse in flow-enforcer, hook error handling (#15)
+- fix: model split always shown, stress inline in footer, not separate line
+- fix: footer uses slot model name, → arrow, inline stress always, remove session-report writes, disable blackbox default
+- fix: sync second footer builder in tool.execute.after with new template
+- fix: compact footer with inline stress gauge, full model names, robust test assertions
+- fix: footer uses trinity tier model name, all 362 tests pass
+- fix: resolve pricing cache corruption, improve TODO extraction, and tune delegation savings
+- fix: use dynamic mcp port fallback
+- fix: handle mcp server close-reopen race
+- fix: await mcp server startup
+- fix: harden prompt send and unblock typecheck
+- fix: sync opencode.json model with brain tier, restore footer icons (trend arrows, stress gauge)
+- fix: deploy script missing vibeOS-api-server/ directory
+- fix: footer prepended to output.output, fix tests, remove stale vibeOS/ directory
+- fix: migrate footer from context-polluting text.complete to UI-only output.title
+- fix: restore experimental.text.complete and message.updated hooks lost during stash
+- fix: ensure model-tiers.json is created when no model is detected
+- fix: update trinity status test for new dashboard format
+- fix: compute cache savings from actual file size, remove /bin/zsh.001 floor, fix state corruption from flow_warns overwrite
+- fix: add proper named export for auto-discovery, fix function closure
+- fix: add startup toast to verify TUI plugin function execution
+- fix: add auto-activation to sync script, add sidebar widget diagnostics
+- fix: restore vibeOS sidebar dashboard widget, fix plugin path in opencode config
+- fix: add size guard to readJsonOrEmpty to prevent OOM on massive state files
+- fix: add generation counter + concurrent-write detection to updateState
+- fix: dedup double footer from competing message.updated / text.complete hooks
+- fix: append ledger entry in recordSaving() and recordCacheSaving()
+- fix: make MCP server close() async, export closeMcpServer for test cleanup
+- fix: isolate tests from real config (chdir sandbox, VIBEOS_MCP_PORT=0, HOME cleanup)
+- fix: release/deploy synced lib deps - blackbox missing caused footer (and all hooks) to disappear
+- fix: resolution-tracker thresholds - isConverging >=0.5, detectLoop Jaccard 0.6, isRefining >-0.01
+- perf: conditional directive injection — skip TDD/FLOW/orchestrator when control vector signals relaxed mode
+- refactor: merge extracted modules into src/index.ts (6656→1061 lines)
+- refactor: extract 16 modules (7207 lines) from src/index.ts into src/lib/
+- refactor: swap blackbox import to LocalBlackboxStub (forensic)
+- refactor: blackbox moved to API-server-only — plugin uses local stub
+- refactor: rename CodeX MCP server to vibeOS MCP server
+- docs: add final stabilization campaign report (#14)
+- docs: add stabilization audit reports for sessions 02-06 and 09 (#13)
+- docs: add stabilization baseline report (#12)
+- docs: update README and AGENTS for remote API protection (Phase 1+2)
+- docs: fix brand name, update AGENTS line count, document shell.env hook
+- docs: update README and AGENTS for v0.9.1 features
+- test: add cross-session restart E2E test (BUG 10)
+- chore: remove TDD auto-generated test artifacts
+- chore: hardcode public VIBEOS_API_TOKEN as default
+- chore: bump to 0.11.0 — blackbox ML engine, loop prevention, pivot detection, API-only architecture
+- chore: replace diagnostic log with visible toast
+- chore: add secrets to .gitignore (.env.production, PRODUCTION-CREDENTIALS.md)
+- ci: add vibeOS test workflow
+- chore: v0.9.1
+bump 0.13.2 — state.ts stub exports, fix ESM import errors
+bump 0.13.1 — trinity optimize (5 modes + auto), compaction every 10 turns, state.ts stub exports
+Merge pull request #18 from DrunkkToys/revert/low-value-api-migration
+revert: undo low-value API migration — scoreStress, extractExports, patterns back to local-only
+Merge pull request #17 from DrunkkToys/feat/remote-api-migration
+test nested
+test api put
+
+
+## 0.13.3
+- feat: blackbox dynamically controls thinking mode per sub-regime for cost savings
+- feat: complete remote API migration — dual-path scoreStress, patternsObserve/Record, TDD exports with local fallback + neutral env test
+- feat: complete remote API migration — dual-path scoreStress, patternsObserve/Record, TDD exports with local fallback
+- feat: blackbox ML enhancements — real features, loop prevention, pivot detection, outcome tracking, calibration
+- feat: v0.10.0 — 6 enhancement phases implemented
+- feat: WordPress integration - atomic seat+token creation
+- feat: Phase 2 - Integrate remote API client into plugin runtime
+- feat: Phase 1 - Remote API server for protected algorithms
+- feat: CodeX MCP server and dashboard sidebar plugin integration
+- feat: vibeOS TUI dashboard sidebar plugin
+- fix: release.mjs — add missing closing brace for deploy else block
+- fix: stabilize refactored modules — ES module bindings, setters, missing imports
+- fix: flow-enforcer race condition, blackbox default ON, dynamic footer
+- fix: lock model name, enforcement logging, TDD framework detection, cache display rounding
+- fix: validateState sessions object, remove stale report writes, drop dead code
+- fix: state validation, flow TODO dedup, session checkpointing, fetch verification
+- fix: _appendFooter full model names, → arrow, inline stress; 361/362 pass
+- fix: atomic state writes, safeJsonParse in flow-enforcer, hook error handling (#15)
+- fix: model split always shown, stress inline in footer, not separate line
+- fix: footer uses slot model name, → arrow, inline stress always, remove session-report writes, disable blackbox default
+- fix: sync second footer builder in tool.execute.after with new template
+- fix: compact footer with inline stress gauge, full model names, robust test assertions
+- fix: footer uses trinity tier model name, all 362 tests pass
+- fix: resolve pricing cache corruption, improve TODO extraction, and tune delegation savings
+- fix: use dynamic mcp port fallback
+- fix: handle mcp server close-reopen race
+- fix: await mcp server startup
+- fix: harden prompt send and unblock typecheck
+- fix: sync opencode.json model with brain tier, restore footer icons (trend arrows, stress gauge)
+- fix: deploy script missing vibeOS-api-server/ directory
+- fix: footer prepended to output.output, fix tests, remove stale vibeOS/ directory
+- fix: migrate footer from context-polluting text.complete to UI-only output.title
+- fix: restore experimental.text.complete and message.updated hooks lost during stash
+- fix: ensure model-tiers.json is created when no model is detected
+- fix: update trinity status test for new dashboard format
+- fix: compute cache savings from actual file size, remove /bin/zsh.001 floor, fix state corruption from flow_warns overwrite
+- fix: add proper named export for auto-discovery, fix function closure
+- fix: add startup toast to verify TUI plugin function execution
+- fix: add auto-activation to sync script, add sidebar widget diagnostics
+- fix: restore vibeOS sidebar dashboard widget, fix plugin path in opencode config
+- fix: add size guard to readJsonOrEmpty to prevent OOM on massive state files
+- fix: add generation counter + concurrent-write detection to updateState
+- fix: dedup double footer from competing message.updated / text.complete hooks
+- fix: append ledger entry in recordSaving() and recordCacheSaving()
+- fix: make MCP server close() async, export closeMcpServer for test cleanup
+- fix: isolate tests from real config (chdir sandbox, VIBEOS_MCP_PORT=0, HOME cleanup)
+- fix: release/deploy synced lib deps - blackbox missing caused footer (and all hooks) to disappear
+- fix: resolution-tracker thresholds - isConverging >=0.5, detectLoop Jaccard 0.6, isRefining >-0.01
+- perf: conditional directive injection — skip TDD/FLOW/orchestrator when control vector signals relaxed mode
+- refactor: merge extracted modules into src/index.ts (6656→1061 lines)
+- refactor: extract 16 modules (7207 lines) from src/index.ts into src/lib/
+- refactor: swap blackbox import to LocalBlackboxStub (forensic)
+- refactor: blackbox moved to API-server-only — plugin uses local stub
+- refactor: rename CodeX MCP server to vibeOS MCP server
+- docs: add final stabilization campaign report (#14)
+- docs: add stabilization audit reports for sessions 02-06 and 09 (#13)
+- docs: add stabilization baseline report (#12)
+- docs: update README and AGENTS for remote API protection (Phase 1+2)
+- docs: fix brand name, update AGENTS line count, document shell.env hook
+- docs: update README and AGENTS for v0.9.1 features
+- test: add cross-session restart E2E test (BUG 10)
+- chore: remove TDD auto-generated test artifacts
+- chore: hardcode public VIBEOS_API_TOKEN as default
+- chore: bump to 0.11.0 — blackbox ML engine, loop prevention, pivot detection, API-only architecture
+- chore: replace diagnostic log with visible toast
+- chore: add secrets to .gitignore (.env.production, PRODUCTION-CREDENTIALS.md)
+- ci: add vibeOS test workflow
+- chore: v0.9.1
+bump 0.13.2 — state.ts stub exports, fix ESM import errors
+bump 0.13.1 — trinity optimize (5 modes + auto), compaction every 10 turns, state.ts stub exports
+Merge pull request #18 from DrunkkToys/revert/low-value-api-migration
+revert: undo low-value API migration — scoreStress, extractExports, patterns back to local-only
+Merge pull request #17 from DrunkkToys/feat/remote-api-migration
+test nested
+test api put
+
+
+## 0.13.2
+
+- fix: add missing stub exports to `state.ts` (computeSavingsPayload, computeStatusPayload, etc.) to resolve ESM import errors in pre-commit hooks and isolated tests
+- fix: sync all compiled `.js` artifacts from `dist-ts/` to `src/lib/` to fix test environment imports
+
+## 0.13.1
+
+- feat: `trinity optimize` command — 5 session-level modes (budget/quality/speed/longrun/auto) with cache-savings-driven auto switching
+- feat: blackbox `OptimizationMode` delta tables — per-mode overrides for all 10 control knobs
+- feat: turn counter + proactive context compaction every 10 turns
+- feat: `autoSelectMode()` — auto switches budget/balanced/quality per sub-regime based on session cache savings
+- feat: footer optimization mode tag (`[BUDGET]`, `[QUALITY]`, `[SPEED]`, `[LONGRUN]`, `[AUTO→BUDGET]`)
+- feat: per-session mode persistence in `blackbox-state.json` (resets to `auto` on restart)
+- feat: mode-specific system prompt directive injection
+- feat: mode locked for session — blackbox CANNOT override user-set mode
+
+## 0.13.0
+
+- refactor: extract 16 modules from src/index.ts into src/lib/ (state, pricing, trinity, TDD, hooks, reporting, research-audit, api-client, credit-api, turn-classify, index-helpers)
+- feat: blackbox dynamically controls thinking mode per sub-regime for cost savings
+- fix: flow-enforcer race condition, blackbox default ON, dynamic footer improvement
+- fix: lock model name, enforcement logging, TDD framework detection, cache display rounding
+- perf: conditional directive injection — skip TDD/FLOW/orchestrator when control vector signals relaxed mode
+- fix: model split always shown, stress inline in footer, not separate line
+- fix: atomic state writes, safeJsonParse in flow-enforcer, hook error handling
+- perf: inline stress in footer, remove session-report writes, disable blackbox default
+- docs: AGENTS.md updated — 8 hooks (added session.compacting), new src/lib/ architecture
+- docs: README updated — added Architecture section with src/lib/ module descriptions
+
+## 0.12.0 (production readiness stabilisation)
+
+- feat: production-ready feature inventory and documentation reconciliation
+- feat: complete test coverage — 362 passing tests, all skeleton files filled
+- fix: error handling — added null-safe DelegationEnforcer default parameter
+- fix: error handling — wrapped all scratchpad I/O in try/catch guards
+- fix: error handling — safeJsonParse adopted in parseJsonc for JSONC tolerance
+- fix: state file integrity — atomic write-then-rename for all 7 state file writers
+- fix: state file integrity — corruption recovery with backup + logging for all 8 readers
+- fix: state file integrity — 10MB size limits prevent OOM on corrupt files
+- fix: MCP server — CORS headers, request logging, input validation, path traversal protection
+- fix: API server — input validation, error handling, SQL injection protection on all 15 route files
+- fix: API server — auth middleware hardening (suspended seat handling, master key auth)
+- fix: scripts — demo_timer.mjs corruption fix, release.mjs syntax fix, sync-ts-build.mjs mapping fix
+- fix: CI/CD — node 20/22 matrix, build step, new release.yml workflow
+- fix: build chain — orphaned dist-ts/ artifacts removed, all TS file mappings verified
+- docs: AGENTS.md — added 7th hook, 8 missing .ts files, 9 missing state files
+- docs: README — added 4 env vars, 3 trinity commands
+
+## 0.11.0
+- feat: per-session model lock (`trinity lock on|off`) — prevents auto-reconcile with OpenCode config changes
+- feat: lock status shown in `trinity status` guards and live footer (`LOCK` tag)
+- feat: blackbox real feature extraction — 11 derived features per turn (word count, question ratio, urgency, sentiment, complexity, etc.)
+- feat: blackbox loop prevention — 4-level escalating intervention (gentle → suggestive → assertive → escalated) injected into system prompts
+- feat: blackbox PIVOT/SWITCH detection — detects context changes outside project scope via drift rate + instruction density change
+- feat: blackbox outcome tracking — detects satisfaction signals from assistant responses (positive/negative/neutral)
+- feat: blackbox online calibration — `POST /api/v1/blackbox/calibrate` auto-tunes thresholds from session outcomes per project
+- feat: blackbox cross-session continuity — project-scoped session keys (not PID), state persists across terminal restarts
+- feat: blackbox API server unification — API server now imports shared ResolutionTracker (removed duplicate implementation)
+- feat: `blackbox_calibration` SQLite table for per-project calibrated weights
+- feat: `blackbox_sessions` now includes outcome column, `/api/v1/blackbox/outcome`, `/api/v1/blackbox/calibration`, `/api/v1/blackbox/project-sessions` endpoints
+- refactor: blackbox moved to API-server-only — plugin uses local stub, full engine runs on server
+- docs: README and AGENTS updated with model lock, blackbox engine, and calibration documentation
+
+## 0.9.3
+- fix: MCP server now starts during DelegationEnforcer init (was trapped in orphaned computeStatusPayload)
+- fix: add mcp_port auto-write to model-tiers.json on init
+- fix: initialize cache_savings_usd in state file on first write
+- fix: deploy.mjs now copies vibeOS-lib/ (including blackbox) and vibeOS-mcp-server.js
+
+## 0.9.2
+- fix: resolution-tracker thresholds — isConverging `> 0.5` → `>= 0.5`
+- fix: resolution-tracker thresholds — detectLoop Jaccard 0.8 → 0.6
+- fix: resolution-tracker thresholds — isRefining `entropyTrend >= 0` → `> -0.01`
+- test: add blackbox evaluation harness with per-regime precision/recall/F1
+
+## 0.9.1
+- feat: vibeOS MCP server HTTP API
+- feat: vibeOS TUI dashboard sidebar plugin
+- chore: sync-ts-build and flow-enforcer enhancements
+
+## 0.9.0
+- feat: stress-mitigation pipeline — detect, warn, harden, reroute
+- fix: TDD skeleton now fires for task subagent file writes
+- fix: smart sub-cent display and 4dp precision for cache savings
+- fix: _refreshModel now reconciles with actual opencode.json model
+- refactor: full decouple — VibeTheOG → vibeOS branding, dirs, reports, test paths
+- docs: add AGENTS.md — immutable project spec for all LLMs
+Merge pull request #1 from DrunkkToys/experiment/process-data-py
+Document pattern learner commands and claimed feature
+Add routine-pattern promotion regression test
+fix telemetry precision and cross-project session linkage
+rename: VibeTheOG → vibeOS
+rename to VibeTheOG and fix OpenCode manifest/plugin path consistency
+release: bump to v0.8.0 and sync stability hardening
+
+
+# Changelog
+
+## 0.8.1
+
+- fix: reduce dummy actions by gating auto TDD skeleton creation to explicit test intent (or direct test-path edits)
+- feat: add active-job persistence and off-topic detection to prompt for scope confirmation before write/edit/task actions
+- chore: patch release bump and docs version sync
+
+## 0.8.0
+
+- chore: minor release bump for stabilized release candidate
+- test: re-validated first-install flow, slot switching, footer integrity, and neutral-environment gates
+
+## 0.7.15
+
+- fix: footer tier refresh now follows active slot changes even when model id is unchanged
+- test: added regression coverage for slot-switch tier updates and stabilized classify fixtures
+- test: neutral-environment validation (`env -i`) re-run before release bump
+
+## 0.7.14
+
+- test: release hardening pass with expanded adversarial coverage (20 tiger-team checks) and full gate validation
+- test: neutral-environment parity validation (`env -i`) confirms typecheck/build/test stability matches baseline
+- test: pre-existing failing write-enforcement test fixed via deterministic model-tiers fixture setup
+- chore: add checkpoint reliability tooling
+  - new `checkpoint-template.md` for structured session handoff
+  - new `scripts/checkpoint-validate.mjs` for schema validation (sections, task IDs/states, diff-stat evidence, handoff checklist)
+  - new `scripts/tests/checkpoint-validate.test.mjs` with pass/fail fixture coverage
+  - new npm scripts: `checkpoint:validate`, `test:scripts`
+- docs: release readiness and SI/cross-session validation workflow standardized for safer iteration
+
+## 0.7.13
+
+- fix: production plugin load hardening — remove hidden `../utils/timer.js` dependency from `session-metrics.js`
+- fix: plugin install payload completeness — include `src/VibeTheOG-lib/session-metrics.js` in desktop sync set
+- fix: runtime config compatibility — remove invalid `plugins` key usage from `opencode.json` workflow assumptions
+- fix: config reader robustness — support both `opencode.json` and `opencode.jsonc` (including JSONC comments/trailing commas)
+- fix: applySlot safety — prefer project-local `opencode.json` to avoid accidental global provider/dropdown mutations
+- test: deep neutral-environment validation (`env -i`) for full suite + build + runtime plugin load (`opencode models deepseek`)
+- docs: README version/install sync updated to match actual runtime dependencies and file paths
+
+## 0.7.12
+
+- chore: TDD strict defaults, flow enforcer improvements, release tooling
+- change: TDD strict mode now defaults to ON (`selection.tdd_strict !== false`)
+- feat: `trinity tdd strict on|off` command to control strict failing TODO templates
+- docs: sample config + README updated for TDD strict defaults and command
+
+## 0.7.10
+
+- feat: compact immutable footer format: `— [model route] | VibeTheOG: X.XX saved ↑|↓|→ —`
+- change: remove noisy footer breakdown segments (flow/tool/rate/duration) from chat footer
+- change: model usage percentages now show only when both brain and worker are actually used
+- test: added footer format contract test to prevent accidental format drift
+
+## 0.7.9
+
+- change: delegation enforcement now defaults to ON (`delegation_enforce !== false`) for safer cost control
+- change: first-run auto-generated `model-tiers.json` now includes `"delegation_enforce": true`
+- docs: sample config updated to show delegation enforcement enabled by default
+
+## 0.7.8
+
+- fix: auto-sync no longer overwrites valid model-tiers.json entries with guessed provider-prefixed IDs
+- fix: provider model IDs now use correct provider prefix (e.g. `deepseek/`) instead of generic `opencode/`
+- fix: task routing skips medium slot when it matches the brain model (fallback to cheap)
+- fix: delegation_enforce defaults to opt-in (`=== true`) instead of opt-out (`!== false`)
+- fix: null-safety guard for enforcement block when no args passed
+- fix: first-install auto-config populates all trinity slots even with single-model fallback
+
+## 0.7.3
+
+- feat: always sync model-tiers.json with opencode.json on every session start (not just first install)
+- feat: detects ALL models from user config — both `provider` dropdown models AND top-level `model` field
+- fix: only writes to model-tiers.json if detected models differ from current config (no unnecessary writes)
+
+## 0.7.2
+
+- feat: auto-create model-tiers.json on first install from opencode desktop provider models (sniffs models from the dropdown menu, no manual config needed)
+- fix: TRINITY_CHEAP/MEDIUM are now mutable so auto-config can refresh them immediately after bootstrap
+
+## 0.6.0
+
+- feat: enhanced live footer with trend indicators (↑↓→), session duration, savings rate/hr
+- feat: per-tool cost breakdown in footer (edit, webfetch, context7, quota, etc.)
+- feat: model usage distribution percentage in footer tag (e.g. [🧠 60% → ⚡ 40%])
+- feat: cache savings displayed as separate line item in footer
+- feat: trend analysis comparing current session rate vs previous sessions
+- perf: extended readLifetimeSavings() to return sesDuration, sesRatePerHour, sesTrend, sesToolBreakdown, sesModelTurns
+- docs: updated README with new footer format documentation
+
+## 0.5.2
+
+- bump: v0.5.2
+
+## 0.5.1
+
+- fix: `trinity thinking full` now stores `"full"` string instead of `null`
+  — status correctly shows "manual" instead of "credit X%"
+  — persists regardless of credit drops (true manual override)
+
+## 0.5.0
+
+- feat: progressive decadence — age-based scratchpad cache rotation
+  (5min fresh → 1h warm/summary → 24h cold → 48h delete, 1000 file / 10MB limit)
+- feat: research audit — scans scratchpad index + session state for WebFetch/WebSearch
+  anti-patterns (domain chains 3+, redundant queries, context7 bypass)
+- feat: reporting framework — persistent reports with consistent schema
+  (auto-saved from research-audit, save/list/read tools, plain-text findings parser)
+- feat: project memory — cross-session continuity via project-states.json
+  (session counter, research patterns, one-shot briefing on fresh session)
+- feat: research-audit, report-save, report-list, report-read tools (5 total)
+- fix: _refreshModel no longer forces currentTier="high" for non-brain slots
+- fix: init tier override only fires for brain slot (not all slots)
+- fix: auto-save moved before early return for totalFetches=0
+- fix: saveReport now auto-parses plain-text findings/metrics (JSON fallback)
+- fix: dedup prevents duplicate auto-saves within 5-minute window
+- fix: TTL prune deletes reports >90d, keeps max 200
+- fix: add ={} to all tool execute signatures to prevent destructuring crash
+- fix: null-guard _wouldBeDuplicate for null summary
+- fix: null-guard summary.slice() in report index update
+- test: deep test (43 pass, 0 fail — module, hooks, routing, project memory)
+- test: mid-tier / brain-slot / credit path test (10/10 pass)
+- test: report framework lifecycle (14/14 pass)
+- test: 4 report fixes (plain-text, dedup, TTL, narrative)
+- test: VENV stress test (57 pass, 2 false-positive from test sequencing)
+- perf: atomic lifetime reads — compute ltTasks+ltCache from single session snapshot
+- perf: getLastLines — 5-line/1024-byte tail replaces fragile 200-byte dedup
+- perf: _readHead — 120-byte header-only read for decadence idempotency checks
+
+## 0.4.9
+
+- chore: bump v0.4.9
+
+## 0.4.8
+
+- fix: scratchpad hash mismatch — stable JSON key sorting matches CC shasum
+- fix: cross-process log dedup — read file tail instead of per-process cache
+- fix: scratchpad inline size cap — auto-prune >2000 files or >20MB
+- fix: clean up savings tag — remove verbose "delegation + cache" breakdown
+
+## 0.4.7
+
+- feat: flow enforcer — lightweight dev-flow rule checks on Write/Edit
+  (fast-path regex, never blocks, dedup per rule+file per session)
+- feat: `trinity flow` / `trinity flow on|off` — audit and toggle
+- test: 9 new flow enforcer unit tests (64 total)
+- fix: log rotation mtime guard to prevent repeated full-file reads
+
+## 0.4.6
+
+- test: 6 new stall-fix tests (system.transform, messages.transform, tool.execute.after)
+- fix: only inject thinking directive when manually set via `trinity thinking`
+- refactor: remove auto credit-based thinking injection (caused stalls)
+
+## 0.4.5
+
+- fix: remove thinking directive from system prompt (caused model stalls)
+- fix: replace imperative "Full content: Read <path>" with neutral cold-storage ref
+- fix: remove Task output compression — brain needs results verbatim
+- fix: only compress webfetch output (HTML/CSS noise)
+
+## 0.4.4
+
+- fix: dedup session-reports.log writes — skip if line unchanged
+- fix: rotate session-reports.log at 500 lines to prevent unbounded growth
+- fix: add pid to log timestamps to distinguish concurrent writers
+- fix: show delegation vs cache breakdown in savings tag
+- fix: prune stale sessions from delegation-state.json (keep latest 30)
+
+## 0.4.2
+
+- bump: version 0.4.2
+- feat: credit API balance fetching (DeepSeek + OpenRouter)
+- fix: thinking level defaults to brief instead of off
+- fix: credit default 50% for sane fallback
+
+
+## 0.3.4
+
+- bump: version 0.3.4
+
+## 0.3.0
+
+- bump: version 0.3.0
+
+## 0.2.4
+
+- bump: version 0.2.4
+
+## 0.2.3b
+
+- bump: version 0.2.3b
+
+## 0.2.3a
+
+- bump: version 0.2.3a
+
+## 0.2.2
+
+- bump: version 0.2.2
+
+## 0.2.1
+
+- bump: version 0.2.1
+
+## 0.2.0
+
+- feat: align alert schema with CC hook; add 4 new tests (39 total)
+- fix: align WRITE_EDIT to $0.07 to match bash hook; fix stale test values
+- feat: port CC features to OC plugin + 35 tests all passing
+
+## 0.1.0
+
+- Initial release — VibeTheOG v3

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 VibeTheOG
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.