auto-coder 3.0.78__tar.gz → 3.0.80__tar.gz

{auto_coder-3.0.78 → auto_coder-3.0.80}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: auto-coder
-Version: 3.0.78
+Version: 3.0.80
 Summary: AutoCoder: AutoCoder
 Author: allwefantasy
 Author-email: allwefantasy <allwefantasy@gmail.com>
@@ -197,8 +197,6 @@ Requires-Dist: GitPython
 Requires-Dist: openai<4,>=3.0.0
 Requires-Dist: anthropic
 Requires-Dist: protobuf
-Requires-Dist: azure-cognitiveservices-speech
-Requires-Dist: real_agent
 Requires-Dist: python-docx
 Requires-Dist: docx2txt
 Requires-Dist: pdf2image

auto_coder-3.0.80/plugins/auto-coder-dream/ui/.auto-coder/backend_sessions.json

@@ -0,0 +1,113 @@
+{
+  "version": 1,
+  "conversations": {
+    "2e285961-dff2-4aeb-83af-ed7dc13a0892": {
+      "conversation_id": "2e285961-dff2-4aeb-83af-ed7dc13a0892",
+      "created_at": 1786884843.931984,
+      "updated_at": 1786885491.014975,
+      "bindings": {
+        "grok": {
+          "backend": "grok",
+          "session_id": "01a00aa2-a381-7772-b4b8-e3d235fa85c5",
+          "updated_at": 1786885491.014977,
+          "metadata": {
+            "source": "external-cli",
+            "rules_prompt_hash": "5503a0b761d427ce575d41d68026cd5a3f7ed4f55730666aebc887e9aaa146f2",
+            "canonical_synced_message_id": "d9315911-5967-4d1b-b4b6-43595a64cbbe"
+          },
+          "created_at": 1786884843.93199
+        }
+      }
+    },
+    "ff053f87-ade5-42d6-8f01-569cb62cab15": {
+      "conversation_id": "ff053f87-ade5-42d6-8f01-569cb62cab15",
+      "created_at": 1786928897.6219132,
+      "updated_at": 1786929522.788262,
+      "bindings": {
+        "grok": {
+          "backend": "grok",
+          "session_id": "01a00d42-d755-7531-a418-7de5f6c2b702",
+          "updated_at": 1786929522.7883499,
+          "metadata": {
+            "source": "external-cli",
+            "rules_prompt_hash": "90f037f365c5e47962e418a07254ba87281b34345a2c36ff092eb7cba3b6f912",
+            "canonical_synced_message_id": "4f9ccdb2-7f42-40d0-9107-af58c77035f4"
+          },
+          "created_at": 1786928897.6219258
+        }
+      }
+    },
+    "15f05fe8-89f4-49a1-8616-00c1eb7a7558": {
+      "conversation_id": "15f05fe8-89f4-49a1-8616-00c1eb7a7558",
+      "created_at": 1786996223.386584,
+      "updated_at": 1786996761.140373,
+      "bindings": {
+        "grok": {
+          "backend": "grok",
+          "session_id": "01a01146-32c9-7143-8ca8-b58566439213",
+          "updated_at": 1786996761.140374,
+          "metadata": {
+            "source": "external-cli",
+            "rules_prompt_hash": "90f037f365c5e47962e418a07254ba87281b34345a2c36ff092eb7cba3b6f912",
+            "canonical_synced_message_id": "c976bd63-33e1-4f6e-b651-f5bdc872038a"
+          },
+          "created_at": 1786996223.386593
+        }
+      }
+    },
+    "b197ae2b-9630-4e45-9c23-6b1920ea9a31": {
+      "conversation_id": "b197ae2b-9630-4e45-9c23-6b1920ea9a31",
+      "created_at": 1787040165.505066,
+      "updated_at": 1787040612.6433,
+      "bindings": {
+        "grok": {
+          "backend": "grok",
+          "session_id": "01a013e4-c7e5-77a1-920b-4ffee10e6266",
+          "updated_at": 1787040612.643302,
+          "metadata": {
+            "source": "external-cli",
+            "rules_prompt_hash": "90f037f365c5e47962e418a07254ba87281b34345a2c36ff092eb7cba3b6f912",
+            "canonical_synced_message_id": "c006c9c4-4727-4f09-8790-e91c80904ae6"
+          },
+          "created_at": 1787040165.505077
+        }
+      }
+    },
+    "b0b4dff9-c820-4d00-8ba1-ccb0648a7af9": {
+      "conversation_id": "b0b4dff9-c820-4d00-8ba1-ccb0648a7af9",
+      "created_at": 1787082628.2652302,
+      "updated_at": 1787083061.688606,
+      "bindings": {
+        "grok": {
+          "backend": "grok",
+          "session_id": "01a0166c-9b8b-7eb2-a9fb-7076e5a355c8",
+          "updated_at": 1787083061.688608,
+          "metadata": {
+            "source": "external-cli",
+            "rules_prompt_hash": "90f037f365c5e47962e418a07254ba87281b34345a2c36ff092eb7cba3b6f912",
+            "canonical_synced_message_id": "978e13c8-ccea-4307-9648-a2262a538519"
+          },
+          "created_at": 1787082628.265239
+        }
+      }
+    },
+    "66c7d199-9d58-49d7-aadf-c95e62d24f7e": {
+      "conversation_id": "66c7d199-9d58-49d7-aadf-c95e62d24f7e",
+      "created_at": 1787126852.227154,
+      "updated_at": 1787127374.696401,
+      "bindings": {
+        "grok": {
+          "backend": "grok",
+          "session_id": "01a0190f-7027-7763-a092-80dd2dc64815",
+          "updated_at": 1787127374.696403,
+          "metadata": {
+            "source": "external-cli",
+            "rules_prompt_hash": "90f037f365c5e47962e418a07254ba87281b34345a2c36ff092eb7cba3b6f912",
+            "canonical_synced_message_id": "c669510f-731d-4eb0-8221-f3be98d9cece"
+          },
+          "created_at": 1787126852.2271638
+        }
+      }
+    }
+  }
+}

auto_coder-3.0.80/plugins/auto-coder-dream/ui/.auto-coder/conversations/conversations/15f05fe8-89f4-49a1-8616-00c1eb7a7558.json

@@ -0,0 +1,305 @@
+{
+  "conversation_id": "15f05fe8-89f4-49a1-8616-00c1eb7a7558",
+  "name": "# Auto-Coder Dream Project Guidance Maintenance\n\nYou maintain the project's long",
+  "description": "# Auto-Coder Dream Project Guidance Maintenance\n\nYou maintain the project's long-lived operating guidance. This is a curation\nand reconciliation task, not a summary task and not a request to maximize ",
+  "created_at": 1786996206.0305011,
+  "updated_at": 1786996761.1342888,
+  "messages": [
+    {
+      "role": "user",
+      "content": "# Auto-Coder Dream Project Guidance Maintenance\n\nYou maintain the project's long-lived operating guidance. This is a curation\nand reconciliation task, not a summary task and not a request to maximize the\nnumber of artifacts. Prefer a small, trustworthy operating manual over broad\ncoverage.\n\nProject path:\n/Users/williamzhu/projects/auto-coder/plugins/auto-coder-dream/ui\n\nSource directories to inspect:\n[\n  {\n    \"kind\": \"auto_coder_conversations\",\n    \"path\": \"/Users/williamzhu/projects/auto-coder/plugins/auto-coder-dream/ui/.auto-coder/conversations\"\n  },\n  {\n    \"kind\": \"auto_coder_logs\",\n    \"path\": \"/Users/williamzhu/projects/auto-coder/plugins/auto-coder-dream/ui/.auto-coder/logs\"\n  },\n  {\n    \"kind\": \"codex_sessions\",\n    \"path\": \"/Users/williamzhu/.codex/sessions\"\n  },\n  {\n    \"kind\": \"codex_archived_sessions\",\n    \"path\": \"/Users/williamzhu/.codex/archived_sessions\"\n  }\n]\n\nExisting dream guidance registry and likely exported guidance locations:\n{\n  \"dream_home\": \"/Users/williamzhu/.auto-coder/dream/users/local\",\n  \"agents_index\": \"/Users/williamzhu/.auto-coder/dream/users/local/AGENTS.md\",\n  \"artifact_roots\": [\n    \"/Users/williamzhu/.auto-coder/dream/users/local/skills\",\n    \"/Users/williamzhu/.auto-coder/dream/users/local/rules\"\n  ],\n  \"guidance_locations\": [\n    {\n      \"kind\": \"dream_agents_index\",\n      \"path\": \"/Users/williamzhu/.auto-coder/dream/users/local/AGENTS.md\",\n      \"exists\": true\n    },\n    {\n      \"kind\": \"dream_skills_dir\",\n      \"path\": \"/Users/williamzhu/.auto-coder/dream/users/local/skills\",\n      \"exists\": true\n    },\n    {\n      \"kind\": \"dream_rules_dir\",\n      \"path\": \"/Users/williamzhu/.auto-coder/dream/users/local/rules\",\n      \"exists\": true\n    },\n    {\n      \"kind\": \"user_codex_agents\",\n      \"path\": \"/Users/williamzhu/.codex/AGENTS.md\",\n      \"exists\": true\n    },\n    {\n      \"kind\": \"user_codex_skills\",\n      \"path\": \"/Users/williamzhu/.codex/skills\",\n      \"exists\": true\n    },\n    {\n      \"kind\": \"user_codex_rules\",\n      \"path\": \"/Users/williamzhu/.codex/rules\",\n      \"exists\": true\n    },\n    {\n      \"kind\": \"user_auto_coder_skills\",\n      \"path\": \"/Users/williamzhu/.auto-coder/.autocoderskills\",\n      \"exists\": true\n    },\n    {\n      \"kind\": \"user_auto_coder_rules\",\n      \"path\": \"/Users/williamzhu/.auto-coder/autocoderrules\",\n      \"exists\": true\n    },\n    {\n      \"kind\": \"project_agents\",\n      \"path\": \"/Users/williamzhu/projects/auto-coder/plugins/auto-coder-dream/ui/AGENTS.md\",\n      \"exists\": false\n    },\n    {\n      \"kind\": \"project_auto_coder_skills\",\n      \"path\": \"/Users/williamzhu/projects/auto-coder/plugins/auto-coder-dream/ui/.auto-coder/.autocoderskills\",\n      \"exists\": false\n    },\n    {\n      \"kind\": \"project_auto_coder_rules\",\n      \"path\": \"/Users/williamzhu/projects/auto-coder/plugins/auto-coder-dream/ui/.auto-coder/autocoderrules\",\n      \"exists\": false\n    },\n    {\n      \"kind\": \"project_codex_skills\",\n      \"path\": \"/Users/williamzhu/projects/auto-coder/plugins/auto-coder-dream/ui/.codex/skills\",\n      \"exists\": false\n    },\n    {\n      \"kind\": \"project_codex_rules\",\n      \"path\": \"/Users/williamzhu/projects/auto-coder/plugins/auto-coder-dream/ui/.codex/rules\",\n      \"exists\": false\n    },\n    {\n      \"kind\": \"project_cursor_rules\",\n      \"path\": \"/Users/williamzhu/projects/auto-coder/plugins/auto-coder-dream/ui/.cursor/rules\",\n      \"exists\": false\n    }\n  ],\n  \"reference_artifacts\": [\n    {\n      \"artifact_id\": \"skill:global:global:pythonmiao-gitcode-release-coordination\",\n      \"kind\": \"skill\",\n      \"scope\": \"global\",\n      \"name\": \"pythonmiao-gitcode-release-coordination\",\n      \"title\": \"PythonMiao GitCode release coordination\",\n      \"description\": \"Build, sign, notarize, install, and smoke one PythonMiao archive, dry-run and publish those exact bytes through the shared GitCode publisher, then verify source tag, remote assets, anonymous download, and SHA-256.\",\n      \"path\": \"/Users/williamzhu/.auto-coder/dream/users/local/skills/global/pythonmiao-gitcode-release-coordination/SKILL.md\",\n      \"project_path\": null,\n      \"generated\": true,\n      \"edited\": false\n    },\n    {\n      \"artifact_id\": \"skill:global:global:posttome-ecosystem-coordination\",\n      \"kind\": \"skill\",\n      \"scope\": \"global\",\n      \"name\": \"posttome-ecosystem-coordination\",\n      \"title\": \"PostToMe ecosystem coordination\",\n      \"description\": \"Route PostToMe API, Go CLI, portable skill, native inbox, and producer changes to their owners, change server contracts first, and verify deployment, release, installation, and app distribution independently.\",\n      \"path\": \"/Users/williamzhu/.auto-coder/dream/users/local/skills/global/posttome-ecosystem-coordination/SKILL.md\",\n      \"project_path\": null,\n      \"generated\": true,\n      \"edited\": false\n    },\n    {\n      \"artifact_id\": \"skill:global:global:auto-coder-chat-client-server\",\n      \"kind\": \"skill\",\n      \"scope\": \"global\",\n      \"name\": \"auto-coder-chat-client-server\",\n      \"title\": \"Auto-Coder client/server coordination\",\n      \"description\": \"Trace canonical and backend-native identity, execution entrypoints, runtime rules, CloudBridge APIs, previews, Dashboard state, model scope, and PythonMiao consumers across the Auto-Coder product family.\",\n      \"path\": \"/Users/williamzhu/.auto-coder/dream/users/local/skills/global/auto-coder-chat-client-server/SKILL.md\",\n      \"project_path\": null,\n      \"generated\": true,\n      \"edited\": false\n    },\n    {\n      \"artifact_id\": \"rule:global:global:llm-cache-usage-ownership\",\n      \"kind\": \"rule\",\n      \"scope\": \"global\",\n      \"name\": \"llm-cache-usage-ownership.md\",\n      \"title\": \"LLM usage, cache, and billing ownership across repositories\",\n      \"description\": \"Normalize provider token and cache semantics once in byzer-llm, consume canonical buckets once in Auto-Coder budgets, pricing, events, and displays, deduplicate external snapshots, and never invent subscription cost.\",\n      \"path\": \"/Users/williamzhu/.auto-coder/dream/users/local/rules/global/llm-cache-usage-ownership.md\",\n      \"project_path\": null,\n      \"generated\": true,\n      \"edited\": false\n    },\n    {\n      \"artifact_id\": \"rule:global:global:global-artifacts-sync-three-locations\",\n      \"kind\": \"rule\",\n      \"scope\": \"global\",\n      \"name\": \"global-artifacts-sync-three-locations.md\",\n      \"title\": \"Global guidance authority and installation surfaces\",\n      \"description\": \"Author durable personal guidance in william-docs, use its conflict-safe Cursor and Auto-Coder linker, and verify Codex, runtime-generated rules, Dream review, and product-owned installers as separate rollout surfaces.\",\n      \"path\": \"/Users/williamzhu/.auto-coder/dream/users/local/rules/global/global-artifacts-sync-three-locations.md\",\n      \"project_path\": null,\n      \"generated\": true,\n      \"edited\": false\n    },\n    {\n      \"artifact_id\": \"rule:global:global:agent-backend-model-no-cross-pollution\",\n      \"kind\": \"rule\",\n      \"scope\": \"global\",\n      \"name\": \"agent-backend-model-no-cross-pollution.md\",\n      \"title\": \"Agent backend/model namespace isolation\",\n      \"description\": \"Resolve model IDs through their owning provider, preserve backend:model pairs across configured, default, conversation, and run scopes, and synchronize verified external catalogs without changing defaults or exposing credentials.\",\n      \"path\": \"/Users/williamzhu/.auto-coder/dream/users/local/rules/global/agent-backend-model-no-cross-pollution.md\",\n      \"project_path\": null,\n      \"generated\": true,\n      \"edited\": false\n    }\n  ]\n}\nThe registry intentionally contains paths and metadata, not document previews.\nRead the full file at a relevant path before deciding whether its guidance is\naccurate, duplicated, or obsolete, but apply the memory-safety boundary below\nbefore reading any large or history-derived file.\n\nExisting generated project artifacts owned by this run. Every artifact_id must\nreceive exactly one keep/update/retire decision in manifest.json:\n<managed_artifacts>\n[\n  {\n    \"artifact_id\": \"rule:project:ui-ddf4a05d:dream-ui-source-and-selectors\",\n    \"kind\": \"rule\",\n    \"name\": \"dream-ui-source-and-selectors.md\",\n    \"title\": \"Dream UI source, selectors, and passive edits\",\n    \"description\": \"Edit only the plugin UI tree, keep all five execution pickers aligned, and send run_immediately=false on Sparse Agent objective PATCH.\",\n    \"path\": \"/Users/williamzhu/.auto-coder/dream/users/local/rules/projects/ui-ddf4a05d/dream-ui-source-and-selectors.md\"\n  },\n  {\n    \"artifact_id\": \"skill:project:ui-ddf4a05d:change-dream-ui\",\n    \"kind\": \"skill\",\n    \"name\": \"change-dream-ui\",\n    \"title\": \"Change Dream UI\",\n    \"description\": \"Change plugin source, keep backend pickers and the packaged web copy in sync, then verify with npm test, npm run build, and a real browser on :4173 or the daemon port.\",\n    \"path\": \"/Users/williamzhu/.auto-coder/dream/users/local/skills/projects/ui-ddf4a05d/change-dream-ui/SKILL.md\"\n  }\n]\n</managed_artifacts>\n\nDiscovery limits already requested by the caller:\n- auto-coder conversations: 200\n- auto-coder logs: 200\n- Codex sessions: 200\n- discovered matching snapshots before agent run: 1\n\nRecent-history memory safety boundary (mandatory):\n- This is an incremental scan. By default inspect only files with\n  `mtime >= 2026-08-16T19:49:59.144266+00:00` (the most recent 24 hours), and within those files\n  keep only messages/events at or after that cutoff. Metadata-only path\n  discovery may be broader; reading file content may not.\n- Before opening any conversation, log, article-like evidence, or transcript,\n  call `stat`. Skip content reads for files larger than 100 MiB and stop\n  after 100 MiB of raw history input across this run. Keep at most\n  1000 matching records.\n- JSONL must be processed as a bounded stream. Read at most\n  100 MiB for one logical record, discard the remainder of an\n  oversized record, parse one\n  record at a time, and keep up to 16 MiB of useful extracted text\n  per selected session. Never accumulate all records or full serialized events.\n- Never use `Path.read_text()`, `Path.read_bytes()`, `open(...).read()`,\n  `.splitlines()`, `list(file_handle)`, or full-event `json.dumps()` on history\n  files. `read_bytes()[:N]` is also forbidden because it reads the whole file\n  before slicing.\n- Do not blanket-skip `compacted` events, tool/function calls or outputs,\n  reasoning, images, or other content-bearing records. They can contain the\n  decisive evidence. Preserve their useful text, structure, filenames, URLs,\n  media types, sizes, and surrounding event context within the same bounds.\n- Do not paste large data URLs or base64 bytes into the model context. Preserve\n  the reference and metadata. When an inline image is relevant, decode only\n  that selected image into `/Users/williamzhu/.auto-coder/dream/users/local/runs/1ad48943-d3b3-45fe-a618-43313976e054/history-assets` with bounded streaming, then inspect\n  the resulting image file; this is evidence preservation, not image omission.\n- Prefer metadata filtering followed by targeted `rg`; cap it with `-n`,\n  `-m 1000`, and `--max-columns 2000` on the\n  bounded candidate list. Do not\n  recursively load every JSONL into Python merely to find which file matches\n  the project.\n- Read older or oversized history only when a recent source names one exact\n  file and the older context is essential. In that exceptional case, use a\n  targeted streaming search with the same record and output bounds; never\n  perform a broad historical backfill.\n- If the recent bounded window contains no durable new evidence, keep verified\n  existing guidance and finish. Do not scan older history just to manufacture\n  an update.\n\nEvidence order and reading workflow:\n1. Read the repository's own `AGENTS.md` completely when it exists. Treat it as\n   the best style and content reference: concrete entrypoints, commands,\n   environment boundaries, completion criteria, and safety rules. Do not\n   replace a valuable detailed manual with a short generated digest.\n2. Read current code, tests, scripts, and authoritative docs to verify every\n   instruction. Current repository evidence outranks generated Dream content.\n3. Audit every managed artifact above and relevant manual/exported guidance.\n   Read full curated guidance files when they pass the size check; this does\n   not authorize full-file reads of conversations, logs, or session JSONL.\n4. Use conversations and Codex/Cursor activity to find repeated corrections,\n   missing procedures, and costly failure modes. They are supporting evidence,\n   not a changelog to reproduce. Prefer project-matching sessions and newer\n   corrections when sources conflict.\n\nContextual concept recovery:\n- User language is often shorthand rather than a file/class name. When a user\n  says something like \"persist the TUI conversation buffer\", read the surrounding\n  conversation to identify which TUI and what \"conversation buffer\" means in\n  that discussion, then trace it through current code, tests, and state files.\n- Start from the newest user-authored messages, not assistant summaries or prior\n  Dream extraction prompts. Find the exact source file and read the messages\n  immediately before and after the phrase. Then inspect `git status` and any\n  current uncommitted files/tests that implement or refine the same concept.\n- Recover the stable concept map: user phrase and useful aliases, owning\n  subsystem/repository, concrete entrypoints, runtime data shape, lifecycle,\n  persistence or API boundary, and invariants that must survive changes.\n- Preserve that verified semantic map when it will help future agents interpret\n  similar requests. Do not preserve the original feature request as a status\n  item after the concept and boundary are understood.\n- Do not discard a concept merely because the user named it once. If current\n  code/tests establish a new durable abstraction or state boundary, capture that\n  stable meaning; discard only the transient implementation status.\n- Put an always-needed term/architecture map in AGENTS.md; use a skill for a\n  repeatable investigation/change workflow; use a rule for a durable invariant.\n  A single concept may need only one of these, not all three.\n\nConcept evidence ledger (required in manifest.json):\n- Write `concept_search_summary` describing which newest user-authored sources\n  you checked, even when no durable shorthand survives review.\n- Write `concepts_reviewed` as an array. For each ambiguous/domain phrase you\n  considered, record `phrase`, exact absolute `source_path`, `context_summary`,\n  recovered `meaning`, owning `owner`, absolute existing `code_paths`,\n  `disposition` (`agents|skill|rule|not_durable`), one exact `destination`\n  (`AGENTS.md`, an emitted relative skill/rule path, or `none`), and `reason`.\n- `source_path` must be a readable regular file, never a directory. For Cursor\n  transcripts, cite the concrete `.jsonl` inside the transcript UUID directory.\n- Every user phrase added to AGENTS.md or an artifact must appear in this ledger.\n  An empty array is valid only after checking the newest user-authored sources.\n- `destination` is the one canonical home for detailed guidance. Other files may\n  link to it in one line but must not repeat its table, lifecycle, or checklist.\n\nPromotion gate — keep or create guidance only when all answers are concrete:\n- Trigger: what future task or condition should activate this guidance?\n- Action: what exact path, command, order, boundary, or check changes behavior?\n- Failure prevented: what plausible wrong action would a capable agent take\n  without it?\n- Evidence: which current repository file, instruction, or repeated correction\n  proves it?\n\nReject and retire:\n- recent-work summaries, WIP lists, dates, commit timelines, release notes, and\n  \"what the user is focusing on\";\n- one-off feature descriptions or resolved bug narratives;\n- generic advice such as read docs, run tests, be careful, or prefer current\n  evidence unless it names the exact project-specific procedure;\n- duplicate fragments that belong in one coherent rule or skill;\n- stale paths, commands, versions, or assumptions contradicted by current code;\n- secrets, credentials, tokens, private excerpts, and copied test-account data.\n\nArtifact split and density:\n- A skill is a reusable multi-step workflow with a clear trigger, prerequisites,\n  ordered commands/actions, expected evidence, failure handling, verification,\n  and a stopping/ask-user condition.\n- A rule is a concise durable invariant or conditional constraint. Consolidate\n  related rules by subsystem instead of creating one file per fact.\n- Do not duplicate repository AGENTS.md verbatim inside a skill/rule. Extract a\n  separate artifact only when conditional loading or procedural depth adds value.\n- Give each fact one canonical home. AGENTS.md may name an artifact and its\n  trigger, but must not repeat the artifact's ordered steps; an artifact must not\n  restate an always-needed AGENTS.md architecture section. Remove the overlap\n  during hostile review.\n- Default budget is 1-6 project artifacts. Zero is valid when repository-owned\n  guidance already covers the useful knowledge.\n\nCreate artifacts under this exact output directory and do not modify the project:\n/Users/williamzhu/.auto-coder/dream/users/local/runs/1ad48943-d3b3-45fe-a618-43313976e054/artifacts\n\nRequired output contract:\n- Write schema_version 2 manifest at `/Users/williamzhu/.auto-coder/dream/users/local/runs/1ad48943-d3b3-45fe-a618-43313976e054/artifacts/manifest.json` with\n  `existing_artifact_decisions`, `artifacts`, `concept_search_summary`, and\n  `concepts_reviewed`.\n- `existing_artifact_decisions` must cover every managed artifact_id exactly\n  once with `decision: keep|update|retire` and a concrete `reason`.\n- keep/update must rewrite the same kind/name and put its `artifact_id` on the\n  emitted manifest item. retire must omit that artifact from files/artifacts.\n- Every emitted artifact needs `kind`, `scope`, `name`, `title`, an operational\n  `description`, relative `path`, and a non-empty `evidence` list. New artifacts\n  omit artifact_id. Use `scope: \"project\"`.\n- Skills go under `/Users/williamzhu/.auto-coder/dream/users/local/runs/1ad48943-d3b3-45fe-a618-43313976e054/artifacts/skills/<skill-name>/SKILL.md`; rules go\n  under `/Users/williamzhu/.auto-coder/dream/users/local/runs/1ad48943-d3b3-45fe-a618-43313976e054/artifacts/rules/<rule-name>.md`. Empty directories are valid.\n- Write an AGENTS.md draft at `/Users/williamzhu/.auto-coder/dream/users/local/runs/1ad48943-d3b3-45fe-a618-43313976e054/artifacts/AGENTS.md` that records\n  a durable repository operating manual, not an artifact inventory.\n- Write a `paths.md` file at `/Users/williamzhu/.auto-coder/dream/users/local/runs/1ad48943-d3b3-45fe-a618-43313976e054/artifacts/paths.md` listing only\n  concrete local directories and important local files mentioned by\n  auto-coder/Codex/Cursor/Claude Code sessions or verified from current\n  project evidence. This file feeds `auto-coder.index`; do not turn it into a\n  full filesystem inventory.\n- In `paths.md`, group entries by source/evidence and use absolute paths in\n  backticks, for example `- /Users/name/projects/repo - mentioned in Codex\n  session cwd`. Include only paths that help future agents navigate.\n- AGENTS.md should preserve valuable detail when complexity requires it. Do not\n  impose an arbitrary bullet limit. Use evidence-driven sections such as\n  architecture/entrypoints, domain vocabulary and state lifecycles, environment\n  boundaries, workflows, verification, completion criteria, and safety.\n- Never add `Recent User Focus`, `Current Project Priorities`, WIP/date/commit\n  summaries, generated-artifact inventories, or links into Dream output paths.\n- Do not mention an artifact merely because it was created. Put detailed\n  conditional procedures in skills/rules and put always-needed operating facts\n  directly in AGENTS.md.\n\nBefore finishing, perform a hostile self-review: remove anything that would not\nchange a future agent's action, verify every command/path against the repository,\nverify all manifest paths exist, and ensure all managed artifact IDs have an\nexplicit decision. A smaller result is a success when it is more trustworthy.",
+      "timestamp": 1786996761.133076,
+      "message_id": "21417865-7ee3-482c-9f31-55472588e7c3",
+      "metadata": {
+        "backend": "grok",
+        "backend_session_id": "01a01146-32c9-7143-8ca8-b58566439213",
+        "source": "external-cli",
+        "turn_kind": "user"
+      }
+    },
+    {
+      "role": "assistant",
+      "content": "Dream UI guidance is still a small operating manual. The last 24 hours of *conversations* were only the prior extractor run; the real new evidence is in the plugin tree after that scan.\n\n**Decisions**\n\n| Artifact | Decision | Why |\n|---|---|---|\n| `dream-ui-source-and-selectors` | **update** | Five pickers, packaged `web/` copy, and `run_immediately: false` on objective PATCH are still true. The Edit Goal & KPI modal now also PATCHes backend, model, and `conversation_policy`. A runtime-only save must not copy `#scheduledAgentRunImmediately` or send `objective`. |\n| `change-dream-ui` | **keep** | Change path is unchanged: edit plugin source only, align pickers + `check.mjs`, `npm test`, `npm run build`, then a real browser on `:4173` or a restarted daemon. |\n\n**What changed in the manual**\n\n- **`conversation_policy`**: `reuse` (default) vs `new`. It is not Reset Conversation (`reset-conversation` / `context_too_large`).\n- **`dsh`**: now required on all five backend selects; `scripts/check.mjs` is the list to trust.\n- Lifecycle stays in `AGENTS.md`: an objective edit on a queued/running service stops that activation and leaves the service paused; runtime-only edits apply on the next naturally prepared activation.\n\nNo extra skills or rules. Plugin-local conversations and the three recent Codex sessions (fusion-agent / deepseek-harness) did not add another durable UI procedure.",
+      "timestamp": 1786996761.13411,
+      "message_id": "c976bd63-33e1-4f6e-b651-f5bdc872038a",
+      "metadata": {
+        "backend": "grok",
+        "backend_session_id": "01a01146-32c9-7143-8ca8-b58566439213",
+        "source": "external-cli",
+        "turn_kind": "user"
+      },
+      "llm_metadata": {
+        "usage_stats": {
+          "usage_schema": "autocoder.usage.v1",
+          "total_input_tokens": 2611063,
+          "regular_input_tokens": 190967,
+          "total_output_tokens": 31071,
+          "regular_output_tokens": 4856,
+          "budget_input_tokens": 190967,
+          "budget_output_tokens": 31071,
+          "provider_total_tokens": 2642134,
+          "input_tokens": 190967,
+          "output_tokens": 31071,
+          "total_tokens": 2642134,
+          "cache_creation_input_tokens": 0,
+          "cache_read_input_tokens": 2420096,
+          "reasoning_tokens": 26215,
+          "billing_mode": "subscription",
+          "cost_available": false,
+          "model_name": "grok-4.6",
+          "request_time": 1786996761.13295,
+          "input_tokens_measured": true,
+          "output_tokens_measured": true,
+          "source": "external:grok"
+        },
+        "cache_info": {
+          "cache_hit": true,
+          "cache_creation_input_tokens": 0,
+          "cache_read_input_tokens": 2420096
+        },
+        "raw_response_metadata": {
+          "source": "external-cli",
+          "backend": "grok",
+          "model": "grok-4.6",
+          "aggregation": "sum_of_authoritative_process_usage",
+          "calls": [
+            {
+              "phase": "session_initialization",
+              "provider_usage": [
+                {
+                  "input_tokens": 42997,
+                  "output_tokens": 62,
+                  "cache_read_input_tokens": 11392,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 57
+                },
+                {
+                  "input_tokens": 42997,
+                  "cache_read_input_tokens": 11392,
+                  "cache_creation_input_tokens": 0,
+                  "output_tokens": 62,
+                  "reasoning_tokens": 57,
+                  "total_tokens": 54451
+                }
+              ],
+              "normalized": {
+                "usage_schema": "autocoder.usage.v1",
+                "total_input_tokens": 54389,
+                "regular_input_tokens": 42997,
+                "cache_read_input_tokens": 11392,
+                "cache_creation_input_tokens": 0,
+                "total_output_tokens": 62,
+                "regular_output_tokens": 5,
+                "reasoning_tokens": 57,
+                "total_tokens": 54451,
+                "budget_input_tokens": 42997,
+                "budget_output_tokens": 62,
+                "input_tokens_measured": true,
+                "output_tokens_measured": true,
+                "provider_total_tokens": 54451,
+                "model_name": "grok-4.6",
+                "source": "external:grok",
+                "input_tokens": 42997,
+                "output_tokens": 62
+              }
+            },
+            {
+              "phase": "agent_turn",
+              "provider_usage": [
+                {
+                  "input_tokens": 48997,
+                  "output_tokens": 674,
+                  "cache_read_input_tokens": 11520,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 229
+                },
+                {
+                  "input_tokens": 7692,
+                  "output_tokens": 304,
+                  "cache_read_input_tokens": 60416,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 39
+                },
+                {
+                  "input_tokens": 3509,
+                  "output_tokens": 592,
+                  "cache_read_input_tokens": 68096,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 293
+                },
+                {
+                  "input_tokens": 13713,
+                  "output_tokens": 301,
+                  "cache_read_input_tokens": 71552,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 298
+                },
+                {
+                  "input_tokens": 2061,
+                  "output_tokens": 801,
+                  "cache_read_input_tokens": 85248,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 798
+                },
+                {
+                  "input_tokens": 6332,
+                  "output_tokens": 1759,
+                  "cache_read_input_tokens": 87296,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 741
+                },
+                {
+                  "input_tokens": 2406,
+                  "output_tokens": 1931,
+                  "cache_read_input_tokens": 93568,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 1928
+                },
+                {
+                  "input_tokens": 2427,
+                  "output_tokens": 2417,
+                  "cache_read_input_tokens": 95872,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 2414
+                },
+                {
+                  "input_tokens": 6202,
+                  "output_tokens": 1393,
+                  "cache_read_input_tokens": 98176,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 662
+                },
+                {
+                  "input_tokens": 5286,
+                  "output_tokens": 1703,
+                  "cache_read_input_tokens": 104320,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 1468
+                },
+                {
+                  "input_tokens": 5436,
+                  "output_tokens": 2145,
+                  "cache_read_input_tokens": 109568,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 2142
+                },
+                {
+                  "input_tokens": 7709,
+                  "output_tokens": 845,
+                  "cache_read_input_tokens": 114944,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 842
+                },
+                {
+                  "input_tokens": 3454,
+                  "output_tokens": 306,
+                  "cache_read_input_tokens": 122624,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 303
+                },
+                {
+                  "input_tokens": 547,
+                  "output_tokens": 267,
+                  "cache_read_input_tokens": 125952,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 264
+                },
+                {
+                  "input_tokens": 11151,
+                  "output_tokens": 1191,
+                  "cache_read_input_tokens": 126464,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 1188
+                },
+                {
+                  "input_tokens": 4352,
+                  "output_tokens": 922,
+                  "cache_read_input_tokens": 137600,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 919
+                },
+                {
+                  "input_tokens": 2025,
+                  "output_tokens": 1374,
+                  "cache_read_input_tokens": 141952,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 1122
+                },
+                {
+                  "input_tokens": 1602,
+                  "output_tokens": 5722,
+                  "cache_read_input_tokens": 143872,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 5719
+                },
+                {
+                  "input_tokens": 6025,
+                  "output_tokens": 3520,
+                  "cache_read_input_tokens": 145408,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 3517
+                },
+                {
+                  "input_tokens": 3584,
+                  "output_tokens": 2193,
+                  "cache_read_input_tokens": 151424,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 973
+                },
+                {
+                  "input_tokens": 2934,
+                  "output_tokens": 290,
+                  "cache_read_input_tokens": 155008,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 287
+                },
+                {
+                  "input_tokens": 526,
+                  "output_tokens": 359,
+                  "cache_read_input_tokens": 157824,
+                  "cache_creation_input_tokens": 0,
+                  "reasoning_tokens": 12
+                },
+                {
+                  "input_tokens": 147970,
+                  "cache_read_input_tokens": 2408704,
+                  "cache_creation_input_tokens": 0,
+                  "output_tokens": 31009,
+                  "reasoning_tokens": 26158,
+                  "total_tokens": 2587683
+                }
+              ],
+              "normalized": {
+                "usage_schema": "autocoder.usage.v1",
+                "total_input_tokens": 2556674,
+                "regular_input_tokens": 147970,
+                "cache_read_input_tokens": 2408704,
+                "cache_creation_input_tokens": 0,
+                "total_output_tokens": 31009,
+                "regular_output_tokens": 4851,
+                "reasoning_tokens": 26158,
+                "total_tokens": 2587683,
+                "budget_input_tokens": 147970,
+                "budget_output_tokens": 31009,
+                "input_tokens_measured": true,
+                "output_tokens_measured": true,
+                "provider_total_tokens": 2587683,
+                "model_name": "grok-4.6",
+                "source": "external:grok",
+                "input_tokens": 147970,
+                "output_tokens": 31009
+              }
+            }
+          ]
+        }
+      }
+    }
+  ],
+  "metadata": {
+    "source": "backend",
+    "backend": "grok",
+    "is_sub_agent": true
+  },
+  "version": 1,
+  "message_count": 2
+}