diff --git a/CHANGELOG.md b/CHANGELOG.md index 6678b52..8d73431 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -2,7 +2,7 @@ ## v3.5.0 (2026-07-22) ### Changed -- **CODEC Pilot parked.** The browser-automation pillar is withdrawn from the product: the Pilot tab is removed from the dashboard and its product card from the Cortex map. Pilot needs a far deeper build than the rest of CODEC to be trustworthy — Google blocks account sign-in from any CDP-controlled browser (not a setting we can change), and cookie walls plus bot challenges make a large share of real sites unusable. The code is parked, not deleted: the skill, route proxy, and PM2 runner were unhooked from the repo and preserved in git history (see `docs/PILOT-PARKED.md`), so it can return when it earns its place. +- **CODEC Pilot parked.** The browser-automation pillar is withdrawn from the product: the Pilot tab is removed from the dashboard and its product card from the Cortex map. Pilot needs a far deeper build than the rest of CODEC to be trustworthy — Google blocks account sign-in from any CDP-controlled browser (not a setting we can change), and cookie walls plus bot challenges make a large share of real sites unusable. The code is parked, not deleted: the skill, route proxy, vendored copy, and PM2 entry were unhooked from this repo, and the engine lives on in the separate `AVADSA25/codec-pilot` repo (see `docs/PILOT-PARKED.md`), so it can return when it earns its place. - **CODEC Project folded into CODEC Overview.** Project mode is a capability of the dashboard rather than a product in its own right. No functionality changes — plan → approve → autonomous run all work exactly as before. - **9 products → 7.** Core · Dictate · Instant · Chat · Vibe · Voice · Overview. @@ -33,7 +33,7 @@ ## v3.1.0 (2026-05-25) ### Security - **Pilot security-hardening wave (PP-1…PP-12)** — full adversarial audit remediation of the browser-automation pillar. AST safety gate at skill-approval time (`skill_review.py` refuses to activate dangerous compiled skills), untrusted-input fencing on the LLM selector-rescue prompt (`replay.py` `build_rescue_prompt` / `wrap_untrusted` so page text can't redirect element selection), irreversible-click blocking on replay unless `PILOT_ALLOW_DESTRUCTIVE=1`, path/glob-traversal neutralization in `slugify()` lookups, and a forensic audit trail (`audit()`) on every skill write/approve/reject/block. -- **`pilot.lucyvpa.com` tunnel removed** from the Cloudflare ingress after an RCE finding — Pilot is local-only again pending re-hardening. +- **Pilot tunnel removed** from the Cloudflare ingress after an RCE finding — Pilot is local-only again pending re-hardening. ### Added - **Conversational continuity in Project mode** — a Project-mode chat thread now binds to the agent it drafts (`_activeAgentId`). Follow-up messages route to that running agent (status queries, mid-run instructions) instead of spawning a duplicate project. Pulsing "Talking to …" chip with one-click exit; auto-clears on terminal status. @@ -73,7 +73,7 @@ - Recent Runs with inline ▶ Replay and 💾 Compile buttons per row - Pending Skills panel with source preview, ✓ Approve / ✕ Reject gate - HITL pause/resume controls on selected run detail -- **Cloudflare tunnel exposure** — `pilot.lucyvpa.com` mapped to `localhost:8094` for off-LAN dashboard access +- **Cloudflare tunnel exposure** — a Pilot hostname mapped to `localhost:8094` for off-LAN dashboard access - **PM2 service `pilot-runner`** added to `ecosystem.config.js` (autorestart, 2GB memory cap, isolated log files) - **Approved skills auto-expose as MCP tools** — generated `pilot_{slug}.py` files carry `SKILL_NAME`/`SKILL_DESCRIPTION`/`SKILL_TAGS` metadata, registered by CODEC's SkillRegistry, callable from voice / Chat / Scheduler / Claude Code / Cursor / VS Code diff --git a/DEMO_SCRIPT.md b/DEMO_SCRIPT.md index 679d06d..ad85857 100644 --- a/DEMO_SCRIPT.md +++ b/DEMO_SCRIPT.md @@ -79,7 +79,7 @@ ## Cut from the demo - **Compare across models** (ex-beat 21) — removed 2026-07-21: needs a licensed machine (AVA cloud) to show 3+ columns; on an unlicensed Mac it shows one and makes no comparison. -- **CODEC Pilot** (ex-beat 13) — parked 2026-07-22, unhooked from production in v3.5. Google blocks account sign-in from any CDP-controlled browser, and cookie walls / bot challenges make a large share of real sites unusable. It needs a far deeper build to be trustworthy; the code is preserved in git history (see `docs/PILOT-PARKED.md`), not deleted. +- **CODEC Pilot** (ex-beat 13) — parked 2026-07-22, unhooked from production in v3.5. Google blocks account sign-in from any CDP-controlled browser, and cookie walls / bot challenges make a large share of real sites unusable. It needs a far deeper build to be trustworthy; the engine is preserved in the separate `codec-pilot` repo (see `docs/PILOT-PARKED.md`), not deleted. - **Live webcam vision** (ex-beat 8) — removed 2026-07-16. The capture was soft (the Anker C200's lens, not code) and it landed as a gadget rather than a capability. Nothing else depends on it. ## Decide before filming diff --git a/FEATURES.md b/FEATURES.md index b4b153c..f601d20 100644 --- a/FEATURES.md +++ b/FEATURES.md @@ -11,19 +11,22 @@ those are concrete engine components rather than separate products. **v2.3 adds Phase 1 (audit + plugin substrate), Phase 2 (continuous observation + automation), Phase 3 (drop-a-project autonomous agents), -Phase 3.5 (UX polish + proactive overlay), and CODEC Pilot (the 8th product -— browser automation you can teach)** — sections 10–14 below. +Phase 3.5 (UX polish + proactive overlay), and CODEC Pilot (then an 8th +product — browser automation you can teach; **parked in v3.5**)** — sections +10–14 below. Phase 3 ships `codec-agent-runner`, the autonomous-agent daemon that makes CODEC a "real AI employee" at the substrate level — drop a project, agent plans + builds + sends updates back proactively, with permission gates and resume-after-restart guarantees throughout. -CODEC Pilot ships the 8th product slot: a dedicated headless Chromium driven -by Qwen, record-by-doing teach mode, deterministic XPath → CSS → LLM-rescue -replay, and an approval gate that protects the SkillRegistry from -prompt-injection-spawned skills. With Project promoted to product #9 in this -release, CODEC is now a **9-product system**. +CODEC Pilot shipped the 8th product slot in v2.3: a dedicated headless Chromium +driven by Qwen, record-by-doing teach mode, deterministic XPath → CSS → +LLM-rescue replay, and an approval gate that protects the SkillRegistry from +prompt-injection-spawned skills. Project was promoted to product #9 in the same +release, making CODEC a 9-product system at the time. **As of v3.5 that is no +longer the shape:** Pilot is parked (§14) and Project folded into CODEC +Overview, so CODEC ships as a **7-product system**. --- @@ -496,7 +499,7 @@ notification dispatch. > challenges make a large share of real sites unusable. The code is preserved in the `codec-pilot` repo. > The inventory below is kept for reference and is **not** counted in CODEC's feature total. -The 8th product — a complete browser-automation pillar with a dedicated headless Chromium, ReAct-style agent loop driven by Qwen, deterministic record-replay with selector fallback, a skill approval gate, and human-in-the-loop takeover. Lives in `~/codec/pilot/` (11 modules), runs as `pilot-runner` on PM2 port 8094. +Formerly the 8th product — a complete browser-automation pillar with a dedicated headless Chromium, ReAct-style agent loop driven by Qwen, deterministic record-replay with selector fallback, a skill approval gate, and human-in-the-loop takeover. The engine now lives in the separate `AVADSA25/codec-pilot` repo (11 modules); its `pilot-runner` PM2 service and `:8094` port were retired from this stack in v3.5. ### Browser substrate (5 features) @@ -573,8 +576,8 @@ The 8th product — a complete browser-automation pillar with a dedicated headle ### Live view + infrastructure (3 features bundled) - **MJPEG live stream** at `/screenshot/stream` — ~3 fps multipart feed, ~350 KB/s; falls back to 2-second polling on disconnect -- **30 HTTP endpoints** on `pilot-runner` (FastAPI, port 8094) with CORS enabled for `codec.lucyvpa.com` cross-origin calls -- **Cloudflare tunnel** at `pilot.lucyvpa.com` for off-LAN dashboard access; PM2 service `pilot-runner` with autorestart + isolated log files +- **30 HTTP endpoints** on `pilot-runner` (FastAPI, port 8094) with CORS enabled for the dashboard origin — *retired in v3.5* +- **Cloudflare tunnel** for off-LAN dashboard access — removed in v3.1 after an RCE finding; the `pilot-runner` PM2 service was retired in v3.5 --- @@ -615,7 +618,7 @@ The 8th product — a complete browser-automation pillar with a dedicated headle **The 9th product + adversarial-hardening release.** CODEC Project promoted to a live product, Pilot security-hardened end-to-end, calmer 2026 palette across the dashboard. - **9 products. One system.** Cortex neural map expanded from 28 → 38 nodes (+31 edges); CODEC Pilot and CODEC Project added as live product cards. -- **Pilot security-hardening wave (PP-1…PP-12)** — full adversarial-audit remediation: AST safety gate at skill-approval time (refuses to activate dangerous compiled skills), untrusted-input fencing on the LLM selector-rescue prompt (page text can't redirect element selection), irreversible-click blocking on replay unless `PILOT_ALLOW_DESTRUCTIVE=1`, path/glob-traversal neutralization in `slugify()` lookups, forensic audit trail on every skill write/approve/reject/block. The `pilot.lucyvpa.com` Cloudflare tunnel was removed after the RCE finding — Pilot is local-only. +- **Pilot security-hardening wave (PP-1…PP-12)** — full adversarial-audit remediation: AST safety gate at skill-approval time (refuses to activate dangerous compiled skills), untrusted-input fencing on the LLM selector-rescue prompt (page text can't redirect element selection), irreversible-click blocking on replay unless `PILOT_ALLOW_DESTRUCTIVE=1`, path/glob-traversal neutralization in `slugify()` lookups, forensic audit trail on every skill write/approve/reject/block. The Pilot Cloudflare tunnel was removed after the RCE finding — Pilot is local-only. - **Conversational continuity in Project mode** — a Project-mode chat thread now binds to the agent it drafts; follow-ups route to the running agent instead of spawning duplicates. Pulsing *"Talking to …"* chip with one-click exit. - **Voice in / voice out** — Kokoro TTS now actually speaks assistant replies when *"Voice Replies"* is on (the toggle was previously inert); per-message Speak button; Pilot tab gets its own 🎤 dictate + 🔊 speak controls. - **Live preview panel** — slide-out shows the most-recently-modified files in a running agent's project folder (5s poll), so you can watch output appear without leaving the chat. diff --git a/README.md b/README.md index a2b257f..4358a7b 100644 --- a/README.md +++ b/README.md @@ -19,7 +19,7 @@ CI GitHub Discussions 370 Features - 76 Skills + 89 Skills 2,000+ Tests 52K+ Lines MIT License @@ -703,7 +703,7 @@ setup_codec.py — Setup wizard (9 steps) ecosystem.config.js — PM2 process management (16 services) ``` -**CODEC Pilot** is parked (v3.5) and no longer part of the product. Its code, skill, route proxy, and PM2 service were unhooked from this repo in v3.5 and preserved in git history (see [docs/PILOT-PARKED.md](docs/PILOT-PARKED.md)). +**CODEC Pilot** is parked (v3.5) and no longer part of the product. Its skill, route proxy, vendored copy, and PM2 entry were unhooked from this repo in v3.5; the engine itself lives on in the separate `AVADSA25/codec-pilot` repo (see [docs/PILOT-PARKED.md](docs/PILOT-PARKED.md)). --- @@ -743,7 +743,7 @@ python3 setup_codec.py > **Project mode** (drop-a-project autonomy) is live and ships as part of **CODEC Overview**. > -> **CODEC Pilot** (browser automation) is **parked as of v3.5.** Google blocks account sign-in from any CDP-controlled browser, and cookie walls plus bot challenges make a large share of real sites unusable — it needs a considerably deeper build before it can be trusted as a product. The code is preserved in git history (see [docs/PILOT-PARKED.md](docs/PILOT-PARKED.md)) rather than deleted. +> **CODEC Pilot** (browser automation) is **parked as of v3.5.** Google blocks account sign-in from any CDP-controlled browser, and cookie walls plus bot challenges make a large share of real sites unusable — it needs a considerably deeper build before it can be trusted as a product. The engine is preserved in the separate `AVADSA25/codec-pilot` repo (see [docs/PILOT-PARKED.md](docs/PILOT-PARKED.md)) rather than deleted. **Phase 3.5 (in progress)** — UX + polish on top of the autonomous-agent substrate: diff --git a/codec_dashboard.py b/codec_dashboard.py index 19f4349..ed89e06 100644 --- a/codec_dashboard.py +++ b/codec_dashboard.py @@ -359,7 +359,7 @@ async def dispatch(self, request, call_next): from routes.vibe_exec import router as vibe_exec_router from routes.web_search import router as web_search_router from routes.cdp import router as cdp_router -# G-series (SR-57..58): cross-source memory search + Pilot proxy. +# G-series (SR-57..58): cross-source memory search. from routes.memory_search import router as memory_search_router from routes.mcp import router as mcp_router # H1 / SR-59: chat handler (POST /api/chat) + its helper cluster. The helpers @@ -817,7 +817,6 @@ async def tasks_page(): return HTMLResponse(f.read(), headers=_NO_CACHE) return HTMLResponse("

Tasks page not found

", status_code=500) -# G2 pilot_proxy → moved to routes/*.py # C3 / SR-38: cortex endpoints moved to routes/cortex.py.