Ideas index: themes and ideas by rank

This issue ranks the themes of ganja's ideas and orders the ideas inside each theme. It was decided in #49, on the map #33.

How to read it

• Rank means priority, not sequence. The next plan is taken from the highest-ranked theme that holds a do idea nothing blocks; themes may run side by side.
• Re-ranking is one edit of this issue. Move a theme or an idea in the list below.
• Inside a theme, a blocker comes before what it blocks; then the source bead's priority; then the smaller size; then the older bead. An idea without a bead comes last unless a dependency places it. The captain broke the two ties the rule left. park and drop ideas are listed after the do ideas and are not ranked.
• At most five do ideas per theme. An idea over the cap is parked until a slot frees up (#36).

Labels

• theme — a theme parent issue; its ideas are its sub-issues.
• idea — an idea about ganja's direction or a new feature. Bugs, tasks and chores are tracked elsewhere.
• verdict:do — the idea will be built. It carries a size: size:S (one commit, no plan), size:M (one plan and one decision number), size:L (several plans, or a new crate or subsystem), size:XL (a new execution environment or an external service).
• verdict:park — kept, with a written condition that reopens it.
• verdict:drop — settled, with a reason; the issue is closed as not planned.

A verdict is a commitment, not an order to start (#36).

Themes by rank

1. #50 — evaluate and Jev judgments

1. #59 Judge seam, [evaluate] table and a marker on untrusted tool results — size M
2. #60 Semantic brake for --auto — size M
3. #61 Self-authored tripwires: a watch tool — size L

park:

• #62 Built-in PreToolUse hook mode for ganja evaluate

2. #51 — Cloud fleet

1. #63 Agents in the cloud with ganja as the control plane — size XL
2. #64 Isolation by default for cloud agents — size XL
3. #65 Scale not bounded by the local machine — size XL

3. #52 — Serve, attach and remote access

1. #66 Name registration for serve-led sessions — size M

park:

• #67 $skills under run --attach

drop:

• #68 Remote Control probe on the claude-code wire

4. #53 — Teammates and team orchestration

1. #69 plan_approval round trip for pane members — size M
2. #70 Read-only or write-capable posture per agent definition — size M
3. #71 Lead notice for a successful shim turn that said nothing — size S
4. #72 codex's queue as the message-injection door for codex panes — size M

park:

• #73 Who may release another member's task claim
• #74 A remove pair for task edges

5. #54 — Cross-session addressing

1. #75 selfSent through the socket peer pid and an ancestry walk — size M
2. #76 Pin persistence across resume — size S
3. #77 Rename grace (formerNames) for peers' pins — size S
4. #78 Pane members as addressable receivers — size M

park:

• #79 A bind opt-out that actually unbinds

6. #55 — MCP and deferred tool loading

1. #80 Skill allowed-tools pre-activates deferred tools — size S
2. #81 The mcp table in its own .mcp.toml — size S
3. #82 Lazy dial of a deferred MCP server — size M
4. #83 Deferred and activated tool counts in /context — size S
5. #84 Semantic reranking of tool_search results — size M

park:

• #85 +term require-mode for tool_search keywords
• #86 Byte-aware deferral sort

7. #56 — Session lifecycle and compaction

1. #87 Reloadable auto_compact_threshold — size S
2. #88 Constraint pinning across compaction — size L

park:

• #89 /restart-handoff command
• #90 compact-plus algorithm for /compact

8. #57 — Provider wires and options

1. #91 Identity the cursor wire presents — size S
2. #92 require_approval for a Responses MCP server tool — size M

park:

• #93 Responses options on the openrouter and compat entries
• #94 Identity ganja presents to GitHub Copilot

9. #58 — Usage, cost and live status display

1. #95 Server-tool call counter in /usage — size S
2. #96 Live per-turn token figure on the working line — size S

park:

• #97 claude-code rate notice in the TUI
• #98 cursor token_details read into Usage

drop:

• #99 Display-only event for a server-picked model

Start by reading this index and its referenced decisions, especially #49, #33, and #36. No source file, test, or entry point is named. Done would require a confirmed change to the ranked themes or ideas, but this issue does not identify one.