Commit Graph
306 Commits
Author SHA1 Message Date
ciregenz a4b3475d3b [eric] browser: flag a wall at 2 repeats and make stuck-nudges diagnose the real cause first 2026-06-04 13:08:25 -07:00
ciregenz a32f53d092 [eric] browser: de-dupe the system prompt and sharpen the terse-output rule 2026-06-04 12:46:49 -07:00
ciregenz 19e7816fc4 [eric] browser: fix stale prompt lines for batch-abort-on-error and smart-wait 2026-06-04 12:18:26 -07:00
ciregenz f3ff27a3ee [eric] browser: on deep stall, nudge a strategic re-plan, not just a selector switch 2026-06-04 12:10:09 -07:00
ciregenz 2768d73a87 [eric] browser: log per-turn output tokens to measure terse-execution wins 2026-06-04 11:43:13 -07:00
ciregenz 3c3dd6cfcc [eric] browser: prompt the agent to plan once then run terse batched turns 2026-06-04 11:37:15 -07:00
ciregenz f59ab9d4dd [eric] auth: drop the mandatory first-launch sign-in gate, app works signed-out 2026-06-04 04:17:31 -07:00
ciregenz 9fd8936cd6 [eric] browser: reset metrics-dir cache per test so suite isolation holds 2026-06-04 04:07:36 -07:00
ciregenz 27686e270b [eric] browser: settle-aware in-batch wait and a terminal read sub-action 2026-06-04 04:04:44 -07:00
ciregenz b77b29408f [eric] browser: cut per-action waste (skip discarded loop-hash, cache metrics dir) 2026-06-03 23:07:45 -07:00
ciregenz a562a5c3c3 [eric] browser: prune stale screenshots + capture downscaled JPEG to cut agent context cost 2026-06-03 22:17:05 -07:00
ciregenz cf8e1f25f7 [eric] websearch: fast-first bounded search/fetch cascade for human-speed 2026-06-03 18:23:08 -07:00
ciregenz 928df4f324 [eric] websearch: gate built-in tool by model route so subscription Claude stops 401ing 2026-06-03 18:15:30 -07:00
ciregenz 33f6e4598e [eric] web: subscription-OAuth WebSearch 401s on token rotation, so keep the reliable DuckDuckGo fallback for those users 2026-06-03 15:29:20 -07:00
ciregenz 3d0dcca22f [eric] browser: per-action command timeouts (reads 12s, nav 20s) so a hung tab surfaces in seconds not 30s 2026-06-03 15:20:37 -07:00
ciregenz bb422c0fe8 [eric] browser: treat a hung tab (repeated timeouts) as card-unavailable so it aborts fast, not a 20-min spin 2026-06-03 15:18:18 -07:00
ciregenz 792767b344 [eric] browser: bound each smart-wait probe so a hung tab can't inherit the 30s command timeout 2026-06-03 15:13:31 -07:00
ciregenz 4f480104dc [eric] browser: route same-flow batch tasks to one sub-agent so BrowserRepeatFlow is reachable 2026-06-03 14:18:00 -07:00
ciregenz 9080f256c9 [eric] browser: log front-loaded perception reads so a read answered from them isn't flagged a ghost 2026-06-03 13:45:34 -07:00
ciregenz 61826d441a [eric] browser: BrowserRepeatFlow hands back capped per-item read data so batch-read actually delivers 2026-06-03 13:05:49 -07:00
ciregenz 50230c88c5 [eric] browser: also sample route count on evaluate (the agent's real read path), not just get_text 2026-06-03 02:11:11 -07:00
ciregenz 7175f4b27b [eric] browser: sample captured-route count on read, not navigate, so the network nudge actually fires 2026-06-03 02:05:11 -07:00
ciregenz 21b225ded4 [eric] browser: nudge the agent toward the fast captured-API tier once per host 2026-06-03 00:50:23 -07:00
ciregenz 186c6cf72e [eric] browser: smart-wait returns when the page's network settles instead of a blind fixed sleep 2026-06-03 00:46:30 -07:00
ciregenz 7572462116 [eric] browser: intra-run batch replay (BrowserRepeatFlow) with verify gate, send-gate, network-route steps 2026-06-03 00:35:48 -07:00
ciregenz 4c0749d0db [eric] browser: subtle Remembered/Learned chip on the agent card so memory feels earned, zero clicks 2026-06-03 00:01:20 -07:00
ciregenz 3a601ba3ad [eric] browser: calm zero-click signals so the user feels the agent recall + learn site memory 2026-06-02 23:46:49 -07:00
ciregenz e6a1a5f2a5 [eric] browser: measure whether the playbook cuts exploration turns, flag when it doesn't 2026-06-02 23:19:47 -07:00
ciregenz f187ec7c6e [eric] browser: surface learned playbook in-chat + REST list/forget so site memory is visible and clearable 2026-06-02 23:16:22 -07:00
ciregenz fe082c492c [eric] browser: durable reflective per-site playbook (tier-2 memory, mem0-style distill+reconcile) 2026-06-02 23:11:47 -07:00
ciregenz 1208a2e746 [eric] browser: don't record a skill when the deliverable was gathered content, replay can't reproduce judgment 2026-06-02 21:20:20 -07:00
ciregenz 78f35aa9b4 [eric] browser: re-check for a replayable skill after the first nav lands on the right host 2026-06-02 19:06:21 -07:00
ciregenz 54d7e3c355 [eric] browser: key skills on the user's stable request, not the orchestrator's reformulation 2026-06-02 18:28:43 -07:00
ciregenz bcfbb8d24f [eric] browser: fail fast with an honest message when the card's webview is gone 2026-06-02 18:19:33 -07:00
ciregenz 34faca12ae [eric] browser: resolve the live card host for replay and report did-nothing runs honestly 2026-06-02 18:14:57 -07:00
ciregenz 68017465ce [eric] browser: add a deterministic completion-honesty gate to catch ghost successes 2026-06-02 18:14:57 -07:00
ciregenz 2b2d5909a8 [eric] browser: instrument skill lifecycle, measure replay speedup, flag silent non-help 2026-06-02 18:03:43 -07:00
ciregenz 294e96d782 [eric] browser: version skills behind a trust gate, compose on proven sub-skills 2026-06-02 18:03:36 -07:00
ciregenz 5bb0b46890 [eric] browser: give the agent tools to list and deprecate its own learned skills 2026-06-02 16:33:21 -07:00
ciregenz e2a77b5eae [eric] browser: parameterize skills so one skill serves a task with different inputs 2026-06-02 16:27:11 -07:00
ciregenz e30d70ce97 [eric] browser: persist skills across sessions with secret-redaction gate 2026-06-02 16:17:23 -07:00
ciregenz 3aa9cf6ba0 [eric] browser: front-load page perception into turn 1 to cut orientation round-trips 2026-06-02 14:02:47 -07:00
ciregenz a015036383 [eric] browser: prompt-cache the fixed prefix and push batching to cut first-run turns 2026-06-02 14:01:00 -07:00
ciregenz 014414440b [eric] browser: decompose BrowserBatch into discrete replayable skill steps 2026-06-02 13:32:54 -07:00
ciregenz d1c1b44ee7 [eric] browser: replay learned skills with zero LLM calls, fall back on any miss 2026-06-02 12:49:55 -07:00
ciregenz d6682190cc [eric] browser: in-memory skill cache that distills tasks into replayable steps 2026-06-02 12:45:42 -07:00
ciregenz 69c4546735 [eric] browser: refine ghost detector so read tasks need returned content, not state change 2026-06-02 12:35:12 -07:00
ciregenz a8964830ac [eric] browser: record per-tool latency, tokens, and per-task completion metrics 2026-06-02 12:12:41 -07:00
ciregenz ea823951c6 [eric] browser: loop test for WebMCP-detect and route list/replay tools 2026-06-02 04:15:08 -07:00
ciregenz 15781d2a3b [eric] browser: list captured routes and replay safe GETs to skip the UI (tier 2) 2026-06-02 04:11:50 -07:00