Commit Graph
7 Commits
Author SHA1 Message Date
rootandClaude Opus 5.5 c437c6bff6 feat(browser): vision automation loop with screenshots, coordinate actions and bounded challenge solving
- browser.screenshot returns the page as an image content block (≤1280px wide, JPEG quality
  configurable) for vision-capable upstreams; non-vision upstreams get a redacted text snapshot.
  Login pages / password fields are marked sensitive and withheld unless a
  browser.sensitive_screenshot grant exists and allowSensitive is passed.
- browser.click_at / type_at / scroll / wait act in screenshot pixels; the server converts via
  viewport/DPR, requires automation control and an allowlisted host, and writes a thumbnail
  evidence file (DATA_DIR/tmp/evidence) referenced from agent_run_event.
- browser.solve_challenge runs a bounded child agent_run (vision caps only, maxSteps ≤ 12,
  budget), judges success from page facts, stops immediately on 2FA / SMS / device confirmation,
  and otherwise moves the session to waiting with evidence and a takeover link.
- PiAgentRuntime passes modelImage outputs as pi-ai image content when model.input has image;
  the ledger never stores image data; coordinate actions use per-invocation idempotency keys.
- Tests: image codec, runtime image content, E14 e2e (login solved, 2FA stop, step limit,
  sensitive withheld). Real camofox 1.17.0 check on example.com recorded in the runbook.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-09-29 01:27:48 +08:00
rootandClaude Opus 5.5 fb84a71f80 style(copy): fix remaining punctuation violations and make check:copy strict
Auto-fix of 602 string literals across 74 files (half-width punctuation next to
Chinese, parenthesised Chinese, quotes); biome reflow; e2/e6 assertions follow
the new full-width summaries. check:copy now runs with --strict in ci.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-09-28 21:34:51 +08:00
rootandClaude Opus 5.5 cf2c4ff0e6 feat(agent): automation-first sessions with step plan, option choice and inline approval
- agent.plan emits plan_updated; system prompt asks for a plan first and updates while executing
- agent.propose_options registers a decision and suspends the run (waitForFact, no worker slot);
  the answer or timeout (recommended/stop) resumes it as the tool result
- console-session runs get inline approvals for plan-bound capabilities: approve replays the call
  with the approval, reject fails it with the reason and records it on the decision
- proposeFirst blocks execute capabilities until an option is chosen; E12 covers all paths

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-09-28 21:07:01 +08:00
rootandClaude Opus 5.5 5c6f483420 feat(agent): agent.spawn{await:true} suspends too; children wake the parent via fact
agent.await and agent.spawn{await:true} now return immediately when
children are still running: the session closes, runAgent registers a
wake signal in agent_run.await_wake and waits with waitForFact without
holding an io/agent worker slot. A child that settles (terminal or
waiting) signals the parent's registered fact, so the parent resumes at
once instead of polling; the suspended call replays and the child
report becomes the tool result. agent.spawn is registered replay-safe;
a suspended spawn is recorded as failed in the ledger so the replay
re-runs it against the same child.

Migration 0037 also adds the admin_recovery_code, admin_passkey and
webauthn_challenge tables used by the account work.

E11 asserts the parent is suspended with no run held by any worker
while it waits, and covers spawn{await:true} end to end.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-09-28 20:21:52 +08:00
rootandClaude Opus 5.5 d64f735810 feat(agent): agent.await suspends the session instead of holding a worker slot
When children are still running, agent.await records the wait, closes
the session (the tool call stays pending in the JSONL) and drive returns
suspended. runAgent waits with waitForFact for the children to settle,
the deadline, or cancellation, then reopens the session; agent.await is
replay-safe, so resume replays it and the model gets the real result.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-09-28 13:45:40 +08:00
rootandClaude Opus 5.5 31a54186c7 copy: second-round copy review across console, settings, agent and server messages
- Trim descriptions, empty states, confirm buttons and toasts to facts
- Unify terms per glossary: 待处理/待识别/整理记录/运行记录/凭证/子会话/Agent 上游
- Subscription-login termsNote states facts only (borrowed client identity, ToS/ban risk)
- Shorter, concrete goal templates for built-in Agent checks
- Human-readable nouns instead of table names in instance CRUD errors
- Audit labels for pi_upstream / schedule / automation_grant actions
- Glossary: add 上游、订阅登录、凭证、定时任务、巡检、运行记录、子会话、停止、持续授权

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-09-28 13:29:42 +08:00
rootandClaude Opus 5.5 459991421e feat(agent): Pi subagents via agent.spawn / agent.await
- agent_run gains depth / spawn_reason (0029); agent_run_event types
  subagent_spawned / subagent_event / subagent_completed
- agent.spawn (propose, medium, replay never): allowlist ⊆ parent, mode not
  above parent, budget carved from parent's remainder, idempotent per run+input;
  settings.pi.subagents {enabled, maxConcurrent, maxDepth} (subagentsEnabled kept as alias)
- child runs claim a concurrency slot under an advisory lock (waitForFact while
  queued); agent.await polls business tables without calling the model
- parent abort cascades to all descendants; parent runtime budget counts
  descendant usage; child events mirrored into the parent stream by parent seq
- authorize caps every child call by all ancestors' allowlists/kinds
- session page shows parent link, inherited info and the subagent tree
- e2e E11; runbook section on subagents

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-09-28 02:49:37 +08:00