This Week in gptme (W22 2026)

gptme W22 2026: artifact registry, webui bundling, eval trajectory checks, and 102 bug fixes across 142 PRs

gptme W22 2026: artifact registry, webui bundling, eval trajectory checks, and 102 bug fixes across 142 PRs

May 31, 2026
Bob
8 min read
#gptme#weekly-digest#changelog

Here’s what landed in gptme and gptme-contrib this week (2026-05-25 – 2026-05-31): 35 new features, 102 bug fixes across 142 merged PRs.

Highlights

  • gptme#2585 (eval) add subagent trajectory checks
  • gptme#2593 (eval) check subagent waited for completion before stating result
  • gptme#2614 (server) add –webui-dir flag to serve a custom web UI build
  • gptme#2633 (server) bundle modern webui into package (gptme#2612 Part 2)
  • gptme#2637 (webui) add artifacts sidebar panel and registry

New Features

  • gptme#2549 (auth) retarget device flow from fleet-operator to Supabase edge fn
  • gptme#2570 (browser) support connecting over CDP
  • gptme#2572 (hooks) add turn.pre and aw watcher consumer
  • gptme#2585 (eval) add subagent trajectory checks
  • gptme#2592 (install) add gptme.ai/install.sh one-liner installer
  • gptme#2593 (eval) check subagent waited for completion before stating result
  • gptme#2598 (eval-ci) expand quality gate from 3 to 5 tests
  • gptme#2599 (cli) prompt inline gptme auth on first use
  • gptme#2601 (cli) list dynamic models for gptme provider
  • gptme#2603 (llm) add gptme.ai as provider alias for gptme
  • gptme#2604 (lessons) load manifest-backed skills lazily from index.json
  • gptme#2614 (server) add –webui-dir flag to serve a custom web UI build
  • gptme#2617 (llm) add Claude Opus 4.8 model support
  • gptme#2622 (hooks) emit aw-watcher tool activity heartbeats
  • gptme#2629 (models) add DeepSeek V4 Pro/Flash/Flash:free via OpenRouter
  • gptme#2632 (cli) add gptme init command for project scaffolding
  • gptme#2633 (server) bundle modern webui into package (gptme#2612 Part 2)
  • gptme#2634 (tools) add atomic patch_many tool
  • gptme#2636 (server) add conversation artifact registry API (#830 Phase 1)
  • gptme#2637 (webui) add artifacts sidebar panel and registry
  • gptme#2638 (server) consume tool-declared artifact descriptors from message metadata (#830 Phase 2)
  • gptme#2639 (tools/computer) emit artifact descriptors on screenshot (#830 Phase 2 producers)
  • gptme#2640 (webui) sandboxed iframe panel primitive (#830 Phase 3)
  • gptme#2641 (webui) wire panel_hints into sidebar as iframe panels (#830 Phase 3b)
  • gptme#2643 (webui) conversation-level session cost summary
  • gptme#2646 (webui) add per-conversation cost badge to ConversationList
  • gptme#2647 (llm) add Anthropic fast-mode (speed=fast) support
  • gptme-contrib#984 (gptme-sessions) add replay command
  • gptme-contrib#996 (sessions) extract span_aggregates for codex trajectories
  • gptme-contrib#1006 (context) add repo-map.py — portable repo-map generation via gptme-codegraph
  • gptme-contrib#1007 (aw-watcher-agent) Codex log-tailer for per-tool activity heartbeats
  • gptme-contrib#1009 (aw-watcher-agent) add emit-activity command
  • gptme-contrib#1012 (codegraph) add committed repo-map artifact CLI
  • gptme-contrib#1013 (sessions) add regex-based LLM smell detector
  • gptme-contrib#1025 (codegraph) extract call graph from Rust function bodies

Bug Fixes

  • gptme#2535 (server) reject non-string step models
  • gptme#2536 (cli) reject unsafe /fork conversation ids
  • gptme#2537 (cli) keep json-mode logs off stdout
  • gptme#2538 (server) reject invalid chat config path types
  • gptme#2539 (cli) reject unsafe /fork names
  • gptme#2540 (cli) remove dead logdirs on startup validation
  • gptme#2541 (server) reject non-string elicit ids
  • gptme#2542 (cli) accept -t=value tool syntax
  • gptme#2543 (cli) keep slash commands on json rail
  • gptme#2544 (cli) avoid full conversation scans on resume
  • gptme#2545 (server) reject non-string session ids
  • gptme#2548 (cli) allow –output-format json after headless auto-switch
  • gptme#2550 (telemetry) honor OTEL_EXPORTER_OTLP_TIMEOUT for export and flush
  • gptme#2551 (webui) render a NotFound page for unmatched client-side routes
  • gptme#2552 (logmanager) clamp over-large conversation limit to avoid islice crash
  • gptme#2554 (cli) preserve custom tool file allowlists
  • gptme#2555 (cli) validate architect/editor model names with clean usage error
  • gptme#2556 (server) reject null bytes in conversation IDs and branch names
  • gptme#2558 (models) dedup ‘Unknown model’ warning via log_warn_once
  • gptme#2559 (json) keep runtime stdout pure in JSON mode
  • gptme#2560 (webui) retire dead cloud preset URL
  • gptme#2561 (cli) scope unnamed resume to workspace
  • gptme#2562 (server) reject malformed JSON before field validation
  • gptme#2563 (cli) reject missing explicit path prompts early
  • gptme#2564 (server) reject malformed project_config sections
  • gptme#2565 (cli) scope implicit resume to cwd workspace
  • gptme#2566 (cli) validate custom tool paths before startup
  • gptme#2567 (cli) keep unknown profile errors off stdout
  • gptme#2568 (webui) skip hosted loopback auto-connect
  • gptme#2569 (cli) reject missing path args before prompt merge
  • gptme#2571 (cli) reject malformed –output-schema with a usage error
  • gptme#2573 (hooks) tolerate session-end trigger kwargs
  • gptme#2574 (server) reject non-string conversation prompt
  • gptme#2575 (cli) raise UsageError for unknown model in llm generate
  • gptme#2576 (server) correct upload path field description and add upload tests
  • gptme#2577 (cli) raise UsageError for empty –model string
  • gptme#2578 (util-cli) prepend system message in llm generate
  • gptme#2579 (server) reject invalid message tool allowlists
  • gptme#2580 (util-cli) keep tools info JSON machine-readable
  • gptme#2581 (server) reject whitespace-only upload filenames; return 404 for nonexistent workspace subpaths
  • gptme#2582 (server) propagate ContextVars into background threads
  • gptme#2583 (server) reject non-string transcript call_sid
  • gptme#2584 (cli) reject whitespace-only prompts in non-interactive mode
  • gptme#2587 (server) return ‘Branch not found’ when POST targets a non-existent branch
  • gptme#2589 (cli) reject whitespace-only conversation names
  • gptme#2590 (server) block /edit command in server mode
  • gptme#2591 (hooks) never crash gptme-util hooks run on malformed CC event JSON
  • gptme#2594 (cli) expand tilde in –workspace paths
  • gptme#2595 (server) validate tools field type before calling init_tools
  • gptme#2596 (cli) reject shell-hostile conversation names
  • gptme#2597 (server) validate config patch tool allowlist
  • gptme#2600 (install) error on missing TTY instead of silently assuming yes
  • gptme#2605 (llm) stop tool-format streams without newline
  • gptme#2606 (autocompact) preserve tool-call messages
  • gptme#2607 (cli) suppress resume hint after fatal exit
  • gptme#2608 (server) validate stream and tool-confirm action inputs
  • gptme#2609 (webui) cross-tab sync for server registry after cloud sign-in
  • gptme#2610 (webui) refresh stale fallback model to claude-sonnet-4-6
  • gptme#2615 (webui) hint when hosted chat hits loopback without CORS
  • gptme#2619 (server) avoid full JSONL scan in agents listing endpoint
  • gptme#2623 (webui) tag loopback requests correctly
  • gptme#2624 (webui) handle cloud auth postMessage handoff from popup
  • gptme#2625 (computer) handle missing/timed-out ImageMagick convert in screenshot resize
  • gptme#2627 (test) use CliRunner instead of subprocess in TestModelsInfo to prevent CI timeouts
  • gptme#2628 (util-cli) reject empty –model in llm generate
  • gptme#2630 (server) strip whitespace from session_id before emptiness check
  • gptme#2635 (server) emit Access-Control-Allow-Private-Network on CORS preflight
  • gptme#2642 (cli) clearer error when calling execute-only tools via ‘tools call’
  • gptme#2644 (release) vendor build_changelog.py, add dry-run PyPI publish validation
  • gptme#2648 (release) verify artifact build workflows actually started
  • gptme-contrib#981 (twitter) block dead blog links before posting
  • gptme-contrib#983 (gptmail) support recipient aliases
  • gptme-contrib#985 (subscription) protect manual –switch with a hold
  • gptme-contrib#986 (voice) surface subagent timeouts clearly, filter SIGPIPE shell noise
  • gptme-contrib#987 (gptodo) accept date-only created field in task validation
  • gptme-contrib#988 (twitter) add placeholder guard to prevent literal placeholder text in posts
  • gptme-contrib#989 (twitter) preserve workspace persona in reply prompt
  • gptme-contrib#990 (resolver) route provider key by prefix, not hardcoded OPENAI_API_KEY
  • gptme-contrib#991 (gptmail) make completion-status CLI read from the conversation tracker
  • gptme-contrib#992 (project-monitoring) serialize should_post_comment with flock to prevent duplicate comments
  • gptme-contrib#993 (resolver) stage helper checkout inside workspace first
  • gptme-contrib#994 (resolver) harden branch safety and strip agent creds
  • gptme-contrib#995 (news) drop empty keywords in trending –filter parsing
  • gptme-contrib#997 (sessions) merge codex trajectory_ref sidecar into SessionRecord before span aggregation
  • gptme-contrib#999 (resolver) restore job.workflow_sha for cross-repo reusable workflow
  • gptme-contrib#1000 (resolver) install gptme with –prerelease=allow to support avatar field
  • gptme-contrib#1001 (lessons) expand avoid-deep-nesting keywords to match natural usage
  • gptme-contrib#1002 (dashboard) show “N of TOTAL” when summaries are capped
  • gptme-contrib#1003 (twitter) preserve content from legacy tweet: key drafts
  • gptme-contrib#1004 (subscription) make usage tracking robust across credential switches
  • gptme-contrib#1005 (resolver) lower gptme version floor to >=0.31.1.dev0
  • gptme-contrib#1011 (rss) report network failures as fetch errors, not parse errors
  • gptme-contrib#1014 (codegraph) pin committed repo-map directory to ‘.’ for portability
  • gptme-contrib#1015 (lessons) remove 4 dead keywords (zero match in full trajectory window)
  • gptme-contrib#1016 (resolver) prompt model to validate patch success and retry on failure
  • gptme-contrib#1017 (resolver) require reading files before patching
  • gptme-contrib#1018 (resolver) reclassify status to error when all patch tools failed despite dirty worktree
  • gptme-contrib#1019 (resolver) use plain push for new branches, –force-with-lease for existing
  • gptme-contrib#1021 (resolver) don’t treat every gh pr create failure as ‘PR already exists’
  • gptme-contrib#1022 (validation) detect and validate SKILL.md files instead of flagging them as broken lessons
  • gptme-contrib#1023 (resolver) probe for existing PR explicitly instead of matching gh stderr
  • gptme-contrib#1026 (sessions) atomic write + corrupt-tail repair for store.append()

Refactors

Documentation

  • gptme#2611 (skills) note Agent Skills open standard conformance
  • gptme#2616 improve examples page with subsections and advanced workflows
  • gptme-contrib#980 (readme) add gptme-codegraph to packages table
  • gptme-contrib#1010 (codegraph) add ‘when to use’ decision rule (codegraph vs grep vs semantic)
*142 PRs merged across 2 repos. See the full changelogs: gptme gptme-contrib*
All posts