Skip to content
AI usage dashboard

Local WorkBuddy / CodeBuddy Code sessions, 2026-09-29

Checked results and limits

  • Official CodeBuddy directories establishes ~/.codebuddy/projects/ session JSONL. Environment variables provides CODEBUDDY_CONFIG_DIR override for configuration/data root. Monitoring also provides OTLP exports, off by default.
  • AgentHUD provider rules/source index reads CodeBuddy .codebuddy/projects and WorkBuddy .workbuddy/projects assistant message/ function_call usage locally, without remote quota reads. tokmesh-core parser/synthetic tests supplies JSONL field/identity samples. aiusage v1.5.8 changes says CodeBuddy message.usage.input_tokens includes cache read, with providerData.rawUsage cache hit/miss; applying Claude-style addition would double count.
  • These latter sources are third-party code/tests, not official product format specifications. Initial inspection found no WorkBuddy session directory; later inspection found native .workbuddy/projects. .codebuddy/projects remains absent. Added native read-only WorkBuddy checks; CodeBuddy CLI still has document-based implementation/synthetic checks only.

Implementation

  • Register independent codebuddy/workbuddy sources with bounded .jsonl discovery in explicit directories. CodeBuddy accepts CODEBUDDY_CONFIG_DIR; manual roots must be projects within the corresponding product directory, avoiding ownership competition between similar parsers.
  • Read completed assistant message/function_call only, requiring valid millisecond timestamps, session/message identity. Missing tokens stay unknown. Invalid numbers/cache read greater than total input/inconsistent hit-miss-input skipped with diagnostics. Key product:session: messageId/traceId/id; line number supplies later-write revision within the same file.
  • input_tokens/prompt_tokens are total input including cache. Explicit prompt_cache_miss_tokens maps to uncached; subtract only with consistent checked fields. reasoning remains an output component. Unverified provider/prices/trace totals are not inferred.
  • Derive normalized total when input/output known and within bounds. Preserve source total separately and compare; inconsistent records excluded from usage.

Validation and remaining conditions

  • cargo test –manifest-path desktop/src-tauri/Cargo.toml -p llm-usage-core tencent_buddy_wire –lib: three synthetic unit tests pass, covering cache deduplication, independent WorkBuddy identity, invalid cache fields, default discovery and incremental cursors.
  • cargo test –manifest-path desktop/src-tauri/Cargo.toml -p llm-usage-core –test adapter_layout_v30: three pass, including separate product directories/registries.
  • Native WorkBuddy last-launch.json reports latest launch 5.6.2, build 37a65c0b33b8e394904eb49e6c218c4cc0601649; this does not identify every historical record’s writer version. Read-only field check: eight JSONL files/420 rawUsage records. All 420 satisfy prompt=cache_miss+cache_hit, total=prompt+completion and message/rawUsage input-output consistency; all message identities unique. Native command: cargo run –manifest-path desktop/src-tauri/Cargo.toml -p llm-usage-core –example real_verify_workbuddy – <home> build/workbuddy-review/real-check-20260929. Eight files/420 events/zero diagnostics; repeat adds zero, stored count stays 420; input=uncached+cache_read and total=input+output pass.
  • Six native ~/.workbuddy/traces/trace_*.json: all trace.totalTokens zero, only two have numeric modelInfo; trace.traceId does not match session providerData.traceId. Without verified duplicate relationships/coverage, trace values are not added to these 420 individual records.
  • Windows 11 x64, Node 22+/Rust lockfile: npm run verify exit 0, Markdown/assets/scripts/UI/ Svelte/fmt/Clippy/Rust/web build. Default-sandbox Node subprocess EPERM retried with automatic approval, original command passed. git diff –check exit 0. Full verification preceded native sample review; later targeted Rust/Clippy/Markdown/fmt checks validated further edits.
  • Obtain native CodeBuddy CLI samples later. WorkBuddy broader versions/subagent-trace overlap/ history after deletion-compaction remain unchecked.
  • CodeBuddy OTLP/local JSONL may represent the same calls; no cross-source duplicate identity or deduplication code yet. Choose one before acceptance. Historical overlap requires source-level resolution, without claiming simple addition establishes complete coverage.

CodeBuddy IDE/plugin local checks and implementation, 2026-09-30

Extension storage checks and limits

  • User reported zero collected CodeBuddy usage. Read-only local ~/.codebuddy contains plugin marketplace/settings/logs; cli-memwatch-* establishes CLI runtime presence, without projects. No local CLI JSONL samples, so original adapter discovered no roots: confirmed cause.
  • IDE candidates have no individual token records: APPDATA/CodeBuddy CN codebuddy-sessions.vscdb ItemTable zero rows; automations.db two empty tables; globalStorage/state.vscdb settings only; genie-history without token fields; workspaceStorage; LOCALAPPDATA/CodeBuddyExtension check-point/file-tree/plan-task.
  • Actual file: LOCALAPPDATA/CodeBuddyExtension/Data/profile/host/workspace/history/session/ conversation/index.json requests[]. Each item has id/type/messages/state/startedAt/usage. Request-level usage aggregates: inputTokens=cacheTokens+cachedMissTokens, totalTokens=inputTokens+outputTokens, plus cachedWriteTokens. lastTokens semantics unverified; credit is platform points, without currency interpretation. Parent session index.json contains only conversations/current registry. messages/id.json extra is embedded JSON string with modelId/modelName/requestId/isHelperMessage. Extracted preparation-permitted fields only, without reading/printing message text.
  • Native sample: 14 conversation indexes, including empty requests. Only one conversation has four type=craft/state=complete requests. One reports input 64339/output 1822/cache read 53504/miss 10835/total 66161, model kimi-k3-1. Three all-zero requests contain a single user message/no model call and are excluded. All four satisfy input=cache+miss and total=input+output. Same session hash occurs across multiple profile/workspace trees, with different conversation IDs; copies may repeat the same request.

Extension implementation

  • codebuddy becomes a combined entry for unchanged CLI JSONL tencent_buddy_wire and new codebuddy/extension_store.rs, dispatched by path shape with shared agent identity. WorkBuddy unchanged, protected by regression.
  • Discovery: one LOCALAPPDATA/CodeBuddyExtension/Data root and bounded conversation-level index.json enumeration; session registries excluded. Manual roots accept that Data directory only.
  • Parse: state!=complete, type!=craft, conflicting cache splits/totals skipped with diagnostics. All-zero usage means no model call and is excluded. Key codebuddy:request:UUID makes cross-tree copies stable within one instance. session_id=conversation ID; host_application= host component, VSCode/CodeBuddyIDE. Bounded reads of messages referenced by requests[].messages choose the last message with model information. Mixed models within one request unverified. credit/lastTokens unmapped. Cursor offset stores file length for unchanged short-circuit; rewrites trigger full replay without duplicate event keys.
  • Added codebuddy-extension-requests-doc1 format registry entry from local checks.

Extension validation and remaining conditions

  • cargo test –manifest-path desktop/src-tauri/Cargo.toml -p llm-usage-core –lib adapters::codebuddy: four pass, request parsing/all-zero/unverified-state skips, duplicate requests within a file, same key across trees, combined dispatch/WorkBuddy protection.
  • Native command: cargo run –manifest-path desktop/src-tauri/Cargo.toml -p llm-usage-core –example real_verify_codebuddy – <home> <localappdata> build/codebuddy-local-verify/work. files=14, added=1, diagnostics=0, one kimi-k3-1 record. Repeat added=0/count stable; input=uncached+cache_read and total=input+output pass, verdict=PASS. Values match native read-only inspection.
  • npm run fmt:check/npm run clippy/npm run test:rust exit 0.
  • CLI projects remains without native samples, document-based only. Extension request aggregate cannot separate multiple model calls within a turn. Mixed-model request attribution and lastTokens/credit semantics need more samples.
  • Full checks rerun at repository root, 2026-09-30: npm run verify exit 0, Markdown/assets/ scripts/UI/Svelte/fmt/Clippy/full Rust/web build; npm run test:browser exit 0, calendar heatmap/ ten languages/themes/timezones/stale responses/user isolation/members/pagination/idle polling. Earlier outstanding full-verification rerun is complete.