Todo¶

Ideas to add to Cogitus

UI¶

Allow key customization (using the Textual built-in functionality and the config file).
continue navigation and keybinding polish.
add contextual dynamic toolbar/actions by mode. [mostly implemented]
add toggle and hotkey for the left tree to expand/collapse all. space does this for the individual nodes, but it would be good to have a toggle that does all (only to the group level, not the root node).
add a tag management dialog with usage counts, rename/edit support, and explicit stale-tag pruning (selected/all stale with confirmation).
revisit the custom footer status widgets and replace the current private textual.widgets._footer.FooterKey dependency with a public Textual API or an internal equivalent once that can be done without reintroducing the past footer layout/refresh regressions. Low priority.

In the db module there are a few places where we still use raw SQL. investigate how we can update sqliter-py to add missing functionality - especially SQL ALTER and listing a helper to list indexes on fields/tables

refactor suitable UI/integration tests to be pilot-first: use Textual's pilot for tests that validate keyboard behavior, focus changes, footer/binding state, screen transitions, or other user-visible interaction flows; keep direct method/event calls only for unit-style helper and branch coverage where pilot would add noise without improving confidence.
review ty type-checker findings in tests, especially whether intentionally invalid SearchFilter construction should use a clearer explicit escape hatch than invalid_field: Any = "status".

upgrade dedicated search results from best-fragment rows to true occurrence-level navigation: let one match row map to one specific occurrence where practical, then scroll the right pane to that exact occurrence and add inline match highlighting there.
optimize advanced search query execution to reduce multi-pass PK collection and re-fetch overhead as dataset size grows.

Add realistic OpenAPI/Swagger response examples to all API endpoints, not just the MCP-exposed read-only routes.

add export to assorted file-types both on the tags, group and idea level.
save and load either entire database or individual ideas
On creating an Idea, allow to use a local or remote file as the seed (ie get a todo from a github repo or local code)
Allow to export an Idea to a remote file (this would be more difficult and need assorted auth additions)

reevaluate the remote API cache design once there is less planned work and technical debt: can we simplify from the current file-backed local cache to a memory-backed cache without losing the needed sync, worker-thread, startup, and cursor-state behavior?
split API database startup initialization from per-request connection creation, then revisit changing synchronous API handlers from async def to def so FastAPI can safely use a threadpool with request-scoped SqliterDB / IdeaService instances.
replace pre-edit full remote sync with a targeted single-idea refresh so editing one idea does not require pulling the whole remote snapshot first.
remove plaintext secret storage from normal app settings: stop persisting the API JWT signing secret in TOML, and move the remote API password to a secret store or runtime-only prompt flow instead of keeping it as plain text in the normal config file.
replace fragile API error-message substring matching with shared typed domain/API exceptions so HTTP 404/409/422 classification does not depend on repository and service error wording.

add full undo ability (basically with git-like functionality, but not using git - a journal that we can unwind?) Include edited, created and deleted ideas at first, more as the need arises. How would we store this?

add profiles, link ideas to a profile, optionally password protected, allow multiple profiles and dialogs to switch between profiles from inside the app.

evaluate optional SQLite write tuning (PRAGMA synchronous=NORMAL) with benchmark and risk assessment.
make idea table writes and FTS index sync failure-tolerant. create(), update(), and delete() currently commit the canonical row change and then update idea_search separately, so an index write failure can leave search stale until rebuild. Investigate a proper fix: shared transaction support if sqliter-py can expose it cleanly, or a transactional outbox/on-commit reindex queue with retry/logging plus an explicit rebuild path.
avoid unconditional FTS index rebuild on every get_db() startup; add an index validity/version check or other lightweight verification so rebuilds only happen when needed, while keeping an explicit/manual rebuild path.

Write a project info or skills file (check how FastAPI does it using https://agentskills.io) so agents can understand how to use the read-only MCP server, when to fall back to CLI JSON export, and what future write support should look like.
Consider a read-only MCP helper that returns full idea details or compact summaries for a query/group in one call, so agents do not need one get_single_idea round trip per result when summarizing many ideas.
Plan future MCP write support as a profile-scoped agent scratchpad: keep broad read-only tools, require write tokens scoped to a Cogitus profile, add explicit JWT claims such as profile and scope, and start with small write tools such as create note, append note, update note, and archive note. Avoid delete/write-anywhere behavior until profile isolation and recovery/history are in place.
explore optional AI-assisted idea expansion and prompt-to-structure workflows.