oceanbase/powercontext

Feature description Add text search to the Dashboard Experience list using the existing family-scoped Artifact search API. Problem and proposed solution The Methods page supports Skill search, but users can only browse Experiences. Reuse the existing search form and Experience search semantics. Keep normal browsing when the query is empty. Acceptance: Queries return matching approved, active Experiences in the selected Scope. Clearing the query restores browsing. Query state is preserved during supported navigation. No matches and request failures have distinct messages. Labels and feedback work in English and Chinese. Add behavior coverage for search, clearing, and failure handling. Additional context The search backend was implemented in #1862; this task only adds Dashboard support. Entry points: src/powercontext/server/dashboard/content.py src/powercontext/server/dashboard/templates/methods.html Coordinate shared Dashboard changes with #1813. This issue will be open for claiming

Feature description Show the existing extraction diagnostics in the human-readable output of powercontext capabilities. Problem and proposed solution The Server exposes extraction configuration, background execution state, and observations, but the CLI currently prints only Memory extraction: enabled/disabled. Display: Configuration status. Background location, role, state, and automatic-processing status. Observation status and window start. Last success and last failure code, stage, and time when available. Handle absent or unknown diagnostics explicitly. Historical observations must not be presented as a current health verdict. Preserve JSON output and use the existing response without additional probes. Add CLI coverage for available, missing, and unverified diagnostics. Additional context The backend diagnostics were implemented in #1796. Entry point: src/powercontext/client/cli.py, _print_human_response. This issue will be open for claiming on October 20, 2026. Please wait unti

Describe the bug The Claude Code status line and Codex token-savings hook can display saved 0 when estimation is disabled or no comparable preparations exist. This makes missing measurements look like measured zero savings. Steps to reproduce Provide a valid scoped statistics response with recall.estimator: null and zero totals. Render it with either integration's formatter. Observe saved 0 instead of an unavailable-estimate state. Expected behavior Distinguish: Estimation disabled. No comparable data in the requested window. Measured zero savings. Positive savings or additional token cost. Valid responses without estimates should remain online. Preserve existing failure classification and accounting semantics. Actual behavior The formatter uses token_reduction without checking whether comparable measurements exist. Environment master ee0589d. Reproduced with controlled formatter inputs. Code: integrations/claude-code/plugins/powercontext/scripts/statusline.py integrations/codex/p

Describe the bug The Codex and Claude Code hooks send the full user prompt as the recall query. Prompts longer than 8,192 characters exceed the PrepareContextRequest limit and cause automatic recall to fail. Steps to reproduce Pass an 8,193-character prompt to either hook's _prepare_context function, capturing the outgoing request. Validate the payload against PrepareContextRequest. Observe a query-length validation error. Expected behavior Use a deterministic bounded query for recall, following the existing ZCode implementation and documenting the truncation policy. Keep Source capture unchanged, including its existing size limit. Add regression coverage for short prompts, the length boundary, and Unicode input. Actual behavior The outgoing query contains the entire prompt and fails validation. Environment master ee0589d. Reproduced through request construction and schema validation, without a native host session. Code: integrations/{codex,claude-code}/plugins/powercontext/hooks/. R

Feature description We have examples as integration with different agents(pi, claude, codex, etc...) I hope to have an example to show how to integrate with pip package like openAI/anthropic. Problem and proposed solution For people and agent to adopt with different agents, a guide or document to show how this repo work with native sdk. Alternatives considered No response Additional context No response Are you willing to contribute to this feature? Yes, I am willing to contribute code, docs, or design feedback.