Skip to content

history.md

Latest commit

 

History

History
613 lines (446 loc) · 25.2 KB

history.md

File metadata and controls

613 lines (446 loc) · 25.2 KB
Raw

Changelog

All notable changes to oh-my-gemini-cli are documented here.

Release Timeline

Version Date Theme Outcome
v0.4.6 2026-03-31 Deep-dive discovery skill Added extension-native $deep-dive trace-to-interview skill with clarity scoring and optional interview/launch state artifacts
v0.4.5 2026-03-30 Deep-interview lock nudge suppression Learn-signal hook now suppresses automated nudges while deep-interview lock is active and resumes safely after lock release
v0.4.4 2026-03-26 Learn-signal safety hardening Added actionable-session filtering, deduped learn-signal state tracking, and safer stale-state handling for /omg:learn nudges
v0.4.3 2026-03-24 AfterAgent deduplication and retry safety Added transcript-fingerprint state tracking so repeated usage-hook retries no longer double-print the same turn
v0.4.2 2026-03-21 Skills/footer compatibility alignment Added slash-friendly omg-plan skill alias and documented Gemini CLI v0.34 skill/footer UX with version-gated guidance
v0.4.1 2026-03-20 Usage monitor runtime knobs Added quiet-hook output control and state-root override for safer, less noisy long sessions
v0.3.9 2026-03-12 Workspace hygiene and hook symmetry Added lane-health auditing, hook lifecycle symmetry rules, and quieter delegated handoffs for safer long sessions
v0.3.8 2026-03-11 Workspace and taskboard control Added extension-native workspace lane mapping and verifier-backed taskboard workflows for lighter long-session orchestration
v0.3.7 2026-03-10 Load surface cleanup Fixed root context wiring, removed mirrored control-plane skills, and slimmed repeated command boilerplate
v0.3.6 2026-03-10 Notification routing Added extension-native notification profiles, event routing, and safe external-delivery boundaries for long-running OmG sessions
v0.3.5 2026-03-04 Dynamic Agent Team Assembly Added approval-gated dynamic team composition with new director/consultant/editor roles and model-aware lane policy
v0.3.4 2026-02-28 Deterministic hook orchestration Added extension-native hook commands/skill with event-lane policy, safety validation, and dry-run testing
v0.3.3 2026-02-28 Claude-style memory management Added file-based memory index/topic split and modular rule-pack workflows
v0.3.2 2026-02-26 HUD visibility controls Added extension-native HUD profile commands/skill and status rendering policy
v0.3.1 2026-02-26 Intent/loop guardrail expansion Added intake gate, loop enforcement, deep init, and conditional rule injection workflows
v0.3.0 2026-02-25 Workflow and mode expansion Added stage pipeline commands, autonomous modes, and lifecycle controls
v0.2.0 2026-02-24 Extensions-first rebuild OmG moved to Gemini CLI's official extension primitives
v0.1.4 2026-02-23 Runtime integration hardening MCP/server wiring and status observability improved
v0.1.3 2026-02-23 Installation path stabilization GitHub-based install flow documented as default path
v0.1.2 2026-02-22 Model/branding consistency gemini-3.1-* naming and OmG branding normalized
v0.1.1 2026-02-22 Dashboard redesign Retro game-style TUI and richer telemetry presentation
v0.1.0 2026-02-22 Initial release Multi-agent orchestration foundation shipped

v0.4.6 - Deep-Dive Discovery Skill (2026-03-31)

Adapted OmG to recent one-week oh-my-claudecode skill direction by adding an extension-native discovery stage that runs trace-first analysis and escalates to interview prompts only when ambiguity remains.

Added

  • New retained skill:
    • skills/deep-dive/SKILL.md ($deep-dive)
  • Structured discovery outputs:
    • clarity score (low/medium/high)
    • assumption ledger for unresolved or provisional decisions
    • launch brief summary for handoff into planning/execution
  • Optional state artifacts for durable handoff:
    • .omg/state/interview-context.json
    • .omg/state/launch-brief.md

Changed

  • README, Korean README, and landing page now document $deep-dive behavior and retained skill count update.
  • Extension/package version bumped to 0.4.6.

Structural Fit Note

  • OmG remains extension-native; no runtime daemon or background worker was introduced.
  • The feature is implemented as prompt/skill/state-file behavior within Gemini CLI extension boundaries.

v0.4.5 - Deep-Interview Lock Nudge Suppression (2026-03-30)

Adapted OmG hook behavior to recent one-week oh-my-codex deep-interview safety direction: when deep-interview lock state is active, automated learn nudges are suppressed so interview flow can continue without interruptions.

Added

  • Deep-interview lock input support in learn-signal flow:
    • read-only lock source: .omg/state/deep-interview.json
  • Additional learn-state visibility fields:
    • deep_interview_lock_active
    • deep_interview_lock_source

Changed

  • hooks/scripts/learn.js now:
    • suppresses nudges when deep-interview lock state is active
    • keeps informational-query filtering, transcript deduplication, and fail-open safety behavior
    • resolves lock-state path through OMG_STATE_ROOT when provided
    • treats recently updated lock snapshots as active for backward-compatible state schemas
  • README, Korean README, and landing page now document deep-interview lock suppression behavior.
  • Extension/package version bumped to 0.4.5.

Structural Fit Note

  • OmG remains extension-native; no runtime daemon or background worker was introduced.
  • The update is hook/state-level only, aligned to existing extension boundaries.

v0.4.4 - Learn-Signal Safety Hardening (2026-03-26)

Hardened the learn-signal hook path so /omg:learn nudges fire only when sessions include actionable implementation intent, while remaining retry-safe under repeated transcript snapshots.

Added

  • New learn-signal state artifact:
    • .omg/state/learn-watch.json
  • New learn-signal hook registration:
    • hooks/hooks.json (AfterAgent -> omg-learn-signal-after-agent)

Changed

  • hooks/scripts/learn.js now:
    • filters informational-only query sessions
    • deduplicates repeated transcript snapshots with a stable event key
    • sanitizes malformed/legacy prior state before reuse
    • keeps fail-open behavior and supports OMG_STATE_ROOT plus OMG_HOOKS_QUIET
  • README, Korean README, and landing page now document learn-signal filtering and state behavior.
  • Extension/package version bumped to 0.4.4.

Structural Fit Note

  • OmG remains extension-native; no runtime daemon or background worker was introduced.
  • Safety hardening stays prompt/state-level and keeps hook operation compact for long sessions.

v0.4.3 - AfterAgent Deduplication and Retry Safety (2026-03-24)

Hardened the built-in AfterAgent usage monitor so repeated hook retries or fallback replays against the same transcript snapshot are treated as already delivered instead of double-printing the same usage line.

Added

  • Transcript fingerprinting for the usage monitor:
    • hooks/scripts/after-agent-usage.js now tracks a stable event key per session/transcript snapshot
    • repeated hook invocations against the same snapshot now short-circuit without emitting duplicate output

Changed

  • hooks/scripts/after-agent-usage.js now persists the last processed transcript fingerprint in .omg/state/quota-watch.json.
  • Usage-monitor state now keeps the retry-safe transcript hash alongside the turn counter and usage snapshot when transcript data is available.
  • README, Korean README, and landing page updated with retry-safe usage-monitor notes.
  • Extension/package version bumped to 0.4.3.

Structural Fit Note

  • OmG remains extension-native; no runtime daemon or background worker was introduced.
  • The change keeps hook retries idempotent without adding new runtime dependencies or delivery channels.

v0.4.2 - Skills and Footer Compatibility Alignment (2026-03-21)

Aligned OmG docs and skill surface with recent Gemini CLI v0.34-era UX updates around slash skill invocation and footer customization, while keeping stable-channel guidance explicit.

Added

  • New slash-friendly planning alias skill:
    • skills/omg-plan/SKILL.md

Changed

  • README, Korean README, and landing page now document:
    • direct skill invocation via /skill-name
    • skill refresh flow via /skills reload
    • footer customization via /footer and ui.footer.* configuration keys
  • Added explicit version gating:
    • stable baseline: Gemini CLI v0.33.0+
    • newer skill/footer UX: Gemini CLI v0.34.0-preview.0+
  • Added collision guidance for native /plan vs OmG planning skill invocation (/omg-plan / $omg-plan).
  • Extension/package version bumped to 0.4.2.

Structural Fit Note

  • OmG remains extension-native; no runtime daemon or background worker was introduced.
  • This release focuses on compatibility clarity and naming-surface safety for newer Gemini CLI UX.

v0.4.1 - Usage Monitor Runtime Knobs (2026-03-20)

Added lightweight runtime controls to the built-in AfterAgent usage monitor so long sessions can reduce noise and redirect state persistence without changing the core extension flow.

Added

  • New usage-monitor runtime environment controls:
    • OMG_HOOKS_QUIET=1 to suppress non-essential hook status output
    • OMG_STATE_ROOT=<dir> to store quota-watch.json outside the default .omg/state path

Changed

  • hooks/scripts/after-agent-usage.js now resolves state storage via OMG_STATE_ROOT when set.
  • Usage monitor state writes are now fail-open to avoid blocking parent workflow execution on local I/O failures.
  • README, Korean README, and hook engineering guide updated with v0.4.1 runtime knob documentation and examples.
  • Extension/package version bumped to 0.4.1.

Structural Fit Note

  • OmG remains extension-native; no new runtime daemon or background process was introduced.
  • Changes are scoped to hook runtime behavior and documentation for safer, quieter operation in long-running sessions.

v0.3.9 - Workspace Hygiene and Hook Symmetry (2026-03-12)

Adapted OmG's extension-side orchestration to recent upstream Gemini CLI changes around dirty worktree safety, hook lifecycle consistency, and subagent policy/handoff clarity without adding runtime daemons or heavier always-on context.

Added

  • New workspace sub-action:
    • /omg:workspace audit
  • New workspace/taskboard state conventions:
    • lane cleanliness/trust/handoff notes inside .omg/state/workspace.json
    • lane-health notes inside .omg/state/taskboard.md
  • New hook lifecycle policy conventions:
    • blocked and early-finish derived hook signals
    • before/after symmetry and continuation re-entry rules in .omg/state/hooks.json

Changed

  • workspace, taskboard, doctor, intent, launch, and status now treat dirty/untrusted lane state as a first-class operational risk.
  • hooks, hooks-init, hooks-validate, and hooks-test now validate lifecycle symmetry and blocked-continuation safety-lane re-entry.
  • team-assemble, team-plan, team-prd, team-exec, team-verify, team, stop, and cancel now keep lane/subagent context explicit while collapsing normal-success verbosity.
  • README, Korean README, and landing page updated with workspace audit, hook symmetry, and Gemini CLI v0.33.0+ compatibility guidance.
  • Extension/package version bumped to 0.3.9.

Structural Fit Note

  • OmG remains extension-native: no runtime daemon, background worker, or terminal hook injector was added.
  • Changes are intentionally prompt/state-level so long sessions stay compact and performance-safe.

v0.3.8 - Workspace and Taskboard Control (2026-03-11)

Added a lightweight control layer inspired by recent worktree-aware planning and stricter completion gates in related harnesses, adapted to OmG's extension-native command/state model.

Added

  • New workspace/taskboard commands:
    • /omg:workspace
    • /omg:taskboard
  • New runtime-state conventions:
    • .omg/state/workspace.json
    • .omg/state/taskboard.md

Changed

  • team-plan now emits stable task IDs, workspace lane hints, and taskboard sync guidance.
  • team-prd, team-exec, and team-verify now align acceptance criteria, implementation slices, and verifier evidence to shared task IDs.
  • team, team-assemble, loop, and autopilot now treat verifier-backed taskboard completion as part of done criteria.
  • launch, status, checkpoint, stop, cancel, cache, optimize, intent, and doctor now recognize workspace/taskboard state.
  • README, Korean README, and landing page updated with workspace/taskboard flows and cache-stability guidance.
  • Extension/package version bumped to 0.3.8.

Structural Fit Note

  • OmG stays extension-native: no new runtime daemon, background worker, or git wrapper was added.
  • Workspace/taskboard state is intentionally compact so long sessions can resume from a stable file anchor instead of replaying verbose chat history.

v0.3.7 - Load Surface Cleanup (2026-03-10)

Fixed the extension manifest/context chain and reduced duplicated discovery surface so OmG loads through a thinner always-on prompt.

Added

  • New root extension context entrypoint:
    • GEMINI.md

Changed

  • GEMINI.md now imports context/omg-core.md, matching the manifest's contextFileName.
  • context/omg-core.md trimmed to stable runtime rules only.
  • Mirrored control-plane skills were removed; retained skills are now:
    • $plan
    • $execute
    • $prd
    • $ralplan
    • $research
    • $context-optimize
  • High-traffic /omg:* commands were shortened by removing repeated output-format boilerplate.
  • doctor diagnostics now validate the real context import chain and retained skill set.
  • Extension/package version bumped to 0.3.7.

Structural Fit Note

  • OmG is now explicitly slash-command-first for operational control.
  • Deep-work skills remain available, but overlapping command/skill mirrors were intentionally removed to shrink startup discovery overhead.

v0.3.6 - Notification Routing (2026-03-10)

Added an extension-native notification policy layer so OmG can surface approvals, verification failures, blockers, checkpoints, and idle drift without depending on custom CLI flags.

Added

  • New notification command and skill:
    • /omg:notify
    • $notify
  • New runtime-state conventions:
    • .omg/state/notify.json
    • .omg/notify/*.md

Changed

  • context/omg-core.md updated with notification controls, persisted state conventions, and worker-session notification safety rails.
  • README, Korean README, and landing page updated with notification profile, channel, and activation guidance.
  • Extension/package version bumped to 0.3.6.

Structural Fit Note

  • OmG notification routing remains extension-native and policy-driven.
  • Actual desktop/webhook delivery is intentionally delegated to Gemini host hooks, shell adapters, or project-specific bridges.

v0.3.5 - Dynamic Agent Team Assembly (2026-03-04)

Added an approval-gated dynamic team composition layer so OmG can assemble fit-for-task rosters before stage execution.

Added

  • New team assembly command and skill:
    • /omg:team-assemble
    • $team-assemble
  • New specialist sub-agents:
    • omg-director
    • omg-consultant
    • omg-editor
  • New guide:
    • docs/guide/agent-team-assembly.md
  • New runtime-state convention:
    • .omg/state/team-assembly.md

Changed

  • team orchestration now supports optional stage-0 dynamic assembly.
  • intent routing expanded with team-assemble classification path.
  • context/omg-core.md updated with stage-0 lifecycle and roster state conventions.
  • README, installation guide, and landing page expanded with team assembly flows, role taxonomy, and model-allocation policy.
  • Extension/package version bumped to 0.3.5.

Structural Fit Note

  • OmG remains extension-native: assembly is prompt/state orchestrated.
  • Team collaboration is modeled through explicit handoffs and verify/fix loops, not hidden runtime daemons.

v0.3.4 - Deterministic Hook Orchestration (2026-02-28)

Added a hook-oriented orchestration layer inspired by additive hook-extension workflows, adapted for Gemini extension-native command/skill/state primitives.

Added

  • New hook command group:
    • /omg:hooks
    • /omg:hooks-init
    • /omg:hooks-validate
    • /omg:hooks-test
  • New hook control skill:
    • $hooks
  • New hook engineering guide:
    • docs/guide/hook-engineering.md
  • New runtime-state artifacts for hook policy and verification:
    • .omg/state/hooks.json
    • .omg/state/hooks-validation.md
    • .omg/state/hooks-last-test.md
    • .omg/hooks/*.md

Changed

  • context/omg-core.md updated with hook controls, hook state conventions, and worker-session side-effect safety rail.
  • README command/skill/docs map expanded with hook workflows.
  • Extension/package version bumped to 0.3.4.

Structural Fit Note

  • OmG hook workflows remain extension-native (prompt/state-driven), not low-level terminal/runtime hook injection.
  • Deterministic lanes and validation/test commands provide predictable behavior before autonomous execution loops.

v0.3.3 - Claude-Style Memory Management (2026-02-28)

Integrated a Claude-style memory model into OmG using a compact root index, topic-split memory files, and path-aware modular rules.

Added

  • New memory command and skill:
    • /omg:memory
    • $memory
  • New memory MCP tools:
    • memory_bootstrap
    • memory_index_sync
    • memory_rules_resolve
    • memory_build_context
  • New file-based memory/rules guide:
    • docs/guide/memory-management.md
  • New reusable memory scaffold assets generated by setup/bootstrap:
    • MEMORY.md
    • .omg/memory/*.md
    • .omg/rules/*.md

Changed

  • omg setup now bootstraps project memory index/topic/rule files.
  • context/omg-core.md updated with memory command and runtime memory artifact conventions.
  • README and landing docs updated to expose memory workflow in command/skill/interface maps.

Structural Fit Note

  • The memory workflow is implemented within extension-native assets and MCP tooling.
  • Rule activation uses frontmatter (description, globs, alwaysApply) for task-conditional loading without bloating base context.

v0.3.2 - HUD Visibility Controls (2026-02-26)

Added extension-native visual status controls inspired by statusline/HUD workflows from related harness projects.

Added

  • New HUD profile commands:
    • /omg:hud
    • /omg:hud-on
    • /omg:hud-compact
    • /omg:hud-off
  • New HUD control skill:
    • $hud
  • New runtime-state artifact for HUD profile persistence:
    • .omg/state/hud.json

Changed

  • /omg:status now renders a HUD section with visibility policy (normal, compact, hidden).
  • context/omg-core.md updated with HUD control conventions.
  • Local dashboard HUD now supports density toggle with h (normal -> compact -> hidden), top summary line, and .omg/state/hud.json sync.
  • Extension/package version bumped to 0.3.2.

Structural Fit Note

  • Gemini Extensions currently support prompt/command/skill/state-driven visual summaries.
  • Direct terminal statusline hook injection remains runtime-specific and outside extension-only primitives.

v0.3.1 - Intent and Loop Guardrail Expansion (2026-02-26)

Added extension-native guardrail workflows inspired by cross-harness operational patterns.

Added

  • New intake/routing command and skill:
    • /omg:intent
    • $intent
  • New loop enforcement command and skill:
    • /omg:loop
    • $loop
  • New deep repository bootstrap command and skill:
    • /omg:deep-init
    • $deep-init
  • New conditional rule injection command and skill:
    • /omg:rules
    • $rules
  • New runtime-state artifacts for extension-guided persistence:
    • .omg/state/intent.md
    • .omg/state/rules.md
    • .omg/state/deep-init.md
    • .omg/state/project-map.md
    • .omg/state/validation.md

Changed

  • context/omg-core.md updated with intent-gating, conditional-rule, deep-init, and loop-discipline conventions.
  • README command/skill map expanded with the new guardrail workflows.
  • Extension/package version bumped to 0.3.1.

Structural Fit Note

  • Extension-manifest primitives can represent policy-oriented orchestration (commands/, skills/, context/) directly.
  • Runtime hook workers, background daemons, and low-level event hooks still require separate runtime code beyond extension-only assets.

v0.3.0 - Workflow and Mode Expansion (2026-02-25)

Added key orchestration capabilities inspired by production usage patterns from related repositories.

Added

  • Stage commands:
    • team-plan
    • team-prd
    • team-exec
    • team-verify
    • team-fix
  • Mode/lifecycle commands:
    • autopilot
    • ralph
    • ultrawork
    • consensus
    • mode
    • launch
    • checkpoint
    • stop
  • New specialist agents:
    • omg-product
    • omg-verifier
    • omg-consensus
  • New workflow skills:
    • $ralplan
    • $prd
    • $autopilot
    • $ralph
    • $ultrawork
    • $consensus
    • $mode
    • $cancel

Changed

  • team orchestration switched to stage-based lifecycle with verify/fix loops.
  • status report expanded to include mode and pipeline stage.
  • Core context updated with mode profiles, lifecycle state conventions, and safety rails.
  • Extension/package version bumped to 0.3.0.

v0.2.0 - Extensions-First Rebuild (2026-02-24)

Reimplemented OmG around Gemini CLI's official Extensions model.

Added

  • Root extension manifest: gemini-extension.json
  • Extension-first package metadata and files list in package.json
  • Sub-agent definitions in agents/
  • Shared extension context in context/omg-core.md
  • Namespaced extension commands in commands/omg/
  • Frontmatter-based skills in skills/

Changed

  • Installation flow migrated to gemini extensions install <repo-url>
  • Documentation rewritten for extension-native usage and command naming
  • Runtime entry shifted from setup scripts to manifest-driven loading

Removed

  • Legacy root-level command templates that depended on !{omg ...} shell execution

Migration Summary

Before After
Global package + setup copier Native extension install via gemini extensions
Script-first runtime bootstrap Gemini extension manifest bootstrap
Mixed onboarding instructions Single extension-first install path

v0.1.4 - Gemini CLI Runtime Integration Hardening (2026-02-23)

Added

  • MCP stdio entry support via @modelcontextprotocol/sdk (--mcp)
  • --server <state|memory|context|orchestrator> server-targeted execution
  • Extended omg status options (--agents, --tasks, --cache, --cache-history, --context, --json)

Changed

  • omg setup updated to register all 4 OmG MCP servers in ~/.gemini/settings.json
  • omg doctor updated to validate all required OmG MCP server registrations
  • Prompt argument wiring adjusted for Gemini CLI compatibility (-p path)
  • Config loading unified via shared config module flow

Impact Matrix

Area Effect
MCP interoperability Better alignment with Gemini CLI runtime expectations
Diagnostics More reliable setup verification and status visibility
Automation Cleaner path for bot-driven execution workflows

v0.1.3 - Installation Path Stabilization (2026-02-23)

Changed

  • GitHub install route promoted as the primary onboarding path
  • npm registry route kept as optional path
  • Gemini CLI package naming corrected in docs (@google/gemini-cli)
  • Troubleshooting section expanded for install-time 404 errors

Tooling

  • Added package lifecycle scripts (prepare, prepack) for build consistency

v0.1.2 - Model and Branding Consistency (2026-02-22)

Changed

  • Model references standardized to gemini-3.1-pro and gemini-3-flash
  • Agent label output updated to display full model names
  • Project brand text normalized from OMG to OmG
  • Tagline changed to: Gemini thinks. OmG orchestrates.

v0.1.1 - Retro Dashboard Refresh (2026-02-22)

Added

  • Retro game-style dashboard skin and status labels (ATK, WIN, KO, ZZZ, RDY)
  • Agent-specific sprite/icon representation in dashboard panels
  • HP/progress bars and party-style member display
  • Quest log and battle log visualization updates

Changed

  • Docs moved under docs/ and linked from repository root README
  • Landing preview updated to match retro dashboard visuals

v0.1.0 - Initial Release (2026-02-22)

Added

  • Core CLI command surface (setup, doctor, team, status, config, bot, update)
  • Multi-agent orchestration primitives:
    • Agent pool and worker lifecycle
    • Task routing and priority queue
    • Agent registry for architect/planner/executor/reviewer/debugger/researcher/quick roles
  • Context-engineering modules:
    • Cache manager
    • Context layering
    • Compaction and prefix optimization
  • MCP server set:
    • omg_state
    • omg_memory
    • omg_context
    • omg_orchestrator
  • Initial docs set (README, guides, landing page)

Compatibility Notes

Component Supported path in this repo
Gemini installation mode gemini extensions install ...
Extension metadata source gemini-extension.json
Command namespace /omg:*
Skill namespace $<skill> from skills/*/SKILL.md (for example: $plan, $team, $intent, $deep-init, $loop, $hud, $memory)

Notes

  • Historical details before v0.1.0 are not tracked.
  • For commit-level details, inspect repository git history.