Skip to content

Triage five remaining README limitations #69

Description

@G-Hensley

This issue is staged for the repository's GitHub tracker cutover. Existing task instructions remain authoritative until the verified handoff.

What

Give each of five README-documented, currently untracked limitations a recorded decision, independent of the already-existing pantheon-08 ticket that owns extending headless dispatch to OpenCode.

Why

The README's "Known gaps and limitations" section groups "headless dispatch supports Claude only" together with "has no live transcript stream" in one bullet, but only the first half is already tracked, by pantheon-08 (extending the shipped Claude-only non-interactive dispatch path to OpenCode). The transcript-stream gap is a separate, still-untracked concern that survives even after pantheon-08 ships, since neither Claude nor OpenCode headless dispatch streams a live transcript today. This item therefore does not duplicate pantheon-08 and cross-references it instead of re-tracking the part it already owns.

Scope

In scope, one independent recorded decision for each of: (1) headless dispatch has no live transcript stream, distinct from and not fixed by pantheon-08; (2) a restored conductor pane gets no briefing until it happens to call list_sessions; (3) the markdown mirror under .pantheon/context/ is write-only and never read back; (4) the fit layout's six-pane ceiling; (5) thin frontend test coverage. Out of scope: fixing any of them as part of this item, and re-tracking headless dispatch's Claude-only scope, which pantheon-08 already owns.

Done when

Each of the five items above has an explicit recorded decision, fix, accept as documented, or revisit later, each stated with the comparison the Validation section below requires, independent of the others.

Validation

For each item, a short bounded comparison against current behavior, the realistic alternative, and the expected cost against the expected benefit: (1) transcript stream: current behavior is a completion notice only; the alternative is streaming partial output to the pane or sidebar; cost is nontrivial plumbing through the existing dispatch queue, benefit is mid-task visibility. (2) conductor briefing: current behavior is silence until the first list_sessions call; the alternative is briefing the pane at MCP connect time; cost is small, benefit is avoiding a restored conductor acting on stale assumptions. (3) markdown mirror: current behavior is write-only; the alternative is making it a read path too, or removing it if nothing consults it; the cost of keeping it write-only is a maintained file nobody reads, the cost of adding a reader is new parsing surface. (4) six-pane ceiling: current behavior hard-caps the fit layout at six panes; the alternative is a scrollable or paginated layout; cost is layout complexity, benefit is supporting larger fleets. (5) frontend coverage: current behavior is thin; the alternative is investing in component and integration tests; cost is ongoing test-authoring time, benefit is regression protection for exactly the UI pieces pantheon-18 and pantheon-11 would add to.

Dependencies and decisions

Each of the five is independent; none blocks another. Item (1) is related to, but not blocked by or blocking, pantheon-08.

Links

README.md, "Known gaps and limitations" section (all five items stated there, the first jointly with the already-owned Claude-only-scope half). pantheon-08 (the existing ticket owning headless dispatch's OpenCode extension, not duplicated here).

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

work-itemEngineering work with explicit scope and acceptance

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions