lrh-work-remains

Imported from xenotaur/logical_robotics_harness (src/lrh/skills/lrh-work-remains/SKILL.md). Install upstream with npx skills add xenotaur/logical_robotics_harness --skill lrh-work-remains. Copyright stays with the author.

lrh-work-remains Skill

This skill answers "what work remains?" for the current session. It is strictly read-only: it never creates, edits, or moves any file, never runs lrh prompt or git mutation commands, and ends at a report — there is no offer-and-write step, unlike this skill's action-oriented siblings (/lrh-closeout, /lrh-confirm-fixes).

Inputs

No argument required. Invoke as:

/lrh-work-remains

Reference Knowledge

Load these before running any step:

1. references/remains-checklist.md — The fixed 18-item checklist this skill reports against, copied verbatim from its source. Do not paraphrase or reorder it.

2. references/grounding-sources.md — Which command(s) ground each checklist category, plus the cross-session-ownership rule (surface and ask, never auto-classify).

Execution Steps

Step 1 — Summarize session accomplishments

Review this session's actual transcript. In one paragraph, state what was accomplished and what prompted it — grounded in what actually happened this session, not a restatement of stated plans that weren't carried out.

Step 2 — Ground each checklist category

For each of the 18 items in references/remains-checklist.md, in order, run the command(s) references/grounding-sources.md lists for that category and report only what the tool output actually shows. If a category has no findings, report that explicitly ("Uncommitted files: none — git status --short is clean") rather than omitting it — a skipped line is indistinguishable from a checked-and-clean one.

Do not substitute conversational memory for a command that could answer the question. If a command isn't applicable in this environment (e.g. gh unavailable, lrh not on PATH, not a git repository), say so explicitly for that category rather than skipping it silently — fall back to the next available signal references/grounding-sources.md lists for that category where one exists (e.g. direct project/ reads when lrh snapshot isn't available).

Step 3 — Flag cross-session ownership candidates

For any branch, PR, or work item surfaced in Step 2 that this session's own transcript never touched, do not report it as this session's own unfinished work and do not silently exclude it either — surface it separately and ask the user to confirm whether it belongs to this session or is already owned by a different one, per references/grounding-sources.md's cross-session ownership rule.

Step 4 — State the next step

If Step 2 surfaced any category with real outstanding work, state the single most logical next step to address it. If everything is clean, say so plainly — do not manufacture a next step where none exists.

Step 5 — Report

Present the Step 1 summary, the full per-category results from Step 2 (all 18 items, including explicit "nothing outstanding" lines), the Step 3 ownership flags (if any), and the Step 4 next step. This is the end of the skill's job — do not offer to act on any finding; if the user wants to act on one, that is a separate, explicit invocation of whichever skill fits.

Quality Checklist

Before reporting completion, verify:

• All 18 checklist categories from references/remains-checklist.md were reported on, in order, none silently omitted
• Each category's claim is grounded in an actual command's output, not conversational recall
• Categories with no findings were reported as "nothing outstanding" explicitly, not left out
• Any branch/PR/WI not touched by this session's own transcript was flagged as a possible cross-session item and asked about, not auto-classified either way
• No file was written, no git/lrh prompt mutation command was run
• A single next step was stated if outstanding work exists, or an explicit "nothing outstanding" if not

What This Skill Does Not Do

• Does not write files, run lrh prompt commands, resolve review threads, or mutate git state in any way — report only.
• Does not automatically classify a branch/PR/WI as belonging to another session — surfaces the candidate and asks.
• Does not offer to act on any finding — that is a separate, explicit invocation of the relevant skill (/lrh-review-response, /lrh-closeout, /lrh-work-item, etc.).
• Does not modify src/lrh/assist/snapshot_cli.py — consumes lrh snapshot current_focus --stdout as-is.
• Does not implement the Taurcode-repo :remains prompt port-back — that lives in the separate Taurcode repo.