Survey output redesign — value & numbers first (folds 8 feedback seeds)#360
Merged
Conversation
…ork mode, manual rename Query-level ACs (AC-1b/AC-3/AC-5/R6#1) for the survey output redesign: - dispatch-fact (za): count distinct in-repo parents that dispatched subagents + total dispatched, via relationship_type='subagent' + parent_session_id parent-join, scoped to the body's exact parent scope. - decision-no-followup (9h): count `done` decisions with no later Edit/Write in the same session, ordered by the real tool_calls.message_id -> messages.ordinal chronological join (not tool_calls.id insertion order). - mode-classification: rename 'mechanical' -> 'manual' for repetitive-but-substantive tracks; add the knowledge-work branch (intake/process/file/log/close loop + content/ops edits, gated on the loop marker, scored beside manual/exploration with the margin guard). - Extend testdata/survey/fixture-sessions.sql: tool_calls.message_id + messages.ordinal, sessions.parent_session_id + relationship_type; seed dispatch (2|3), no-followup (1), knowledge-work (notes-ops), all-unlabeled-Codex mutation. Each expected value is a fixture-derived oracle with a non-vacuous mutation (higher-ordinal Edit decrements no-followup; out-of-repo subagent re-point flips dispatch; kloop-strip drops knowledge-work). 14/14 TestSurveyQuerySmoke subtests pass; 26/26 integration package. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Render-level redesign (AC-1/AC-2/AC-4/AC-6), folding all six feedback seeds into the captain-locked output structure: - step 4 report template rewritten: SpaceDock-survey title carrying the window (5x), recent-window-snapshot + agent-logs-only subhead, WHAT THIS GIVES YOU plain lede (9h), BY THE NUMBERS block (interruptions, hanging-threads count, no-follow-up count, sessions-read with the workdir-vs-name caveat, conditional za dispatch line), HOW YOU WORK de-jargoned loop + the knowledge-work archetype (zb#2), the ↓ full-analysis pointer, then the demoted detail: THREADS TO PULL reframe of the frontier (9h#4), BACKLOG, RECENT DECISIONS, WORKSTREAMS/WORK-BY-AREA (with the branch-and-merge caveat, zw#2)/CODEX (with the conditional all-unlabeled collapse, h5#1)/SCAFFOLD/INTERRUPTIONS, and one consolidated WHAT THIS CAN'T SEE block (5x). - explicit no-scratch-preamble instruction (h5#2): the first line is the title, never "I have everything I need" / "Let me ...". - mode-aware commission offer (zw#1): manual keeps gate-and-drive; exploration leads with iterate/steer; knowledge-work gets batch book-keeping. - step 2 run_query list + prose updated (dispatch-fact, decision-no-followup, knowledge-work archetype, manual rename); Overview + synthesis guidance updated to match. Proven by a LIVE survey drive over this repo's real agentsview corpus (68 sessions, 33 orchestrating parents, decision-no-followup=2, dispatch=33/1720) rendering the new spine with corpus-derived numbers and no scratch preamble — not a SKILL.md prose-grep. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Clarity-only edits (Strunk): split run-ons in the WHAT THIS GIVES YOU lede and the
manual/exploration/knowledge-work offers, tighten 'roughly N times in this window',
'these passed N times'. State-list notation kept in slash form to match the report's
state glyphs. No {slot} tokens or structure changed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…bent scaffold detection (A) Knowledge-work offer reframe (R5/R6): the shipped offer framed knowledge-work as passive book-keeping and disclaimed automation — backwards. Rewrite the step-4 knowledge-work mode-keyed offer to a per-thread tracker that AUTO-RUNS the processing the user already built: long-lived per-entity threads (per recurring counterpart / advisory topic / one-off interview-or-strategy pass); a new entry auto-runs the user's existing processing/commentary skills (their index/update skill becomes ONE automated step), stopping only at the judgment calls (intake/filing + commentary emphasis). Drop the "book-keeping" / "no automation" framing; mirror in the synthesis-guidance mode-keyed-offers bullet. Proven by a live drive over the knowledge-work fixture (notes-ops → knowledge-work; rendered offer names per-entity threads + auto-running existing skills + judgment-call gates, no "no automation"). (B) Spacedock-incumbent scaffold detection (R7; folds seed dyxqywnwb4c3zwb3pka1p6s0): the DB tally excludes WHERE family <> 'spacedock' (stops the survey's own self-call reading every repo as spacedock), which also hid spacedock from SCAFFOLD on a genuine spacedock repo. Add a spacedock FILE-PROBE to step 3 — a runnable spacedock_incumbent() bash function matching .spacedock-state/ / a workflow README with spacedock frontmatter / _mods/ — so SCAFFOLD names spacedock WHEN a workflow is on disk, while KEEPING the tally exclusion. The file-probe distinguishes the genuine incumbent from the self-call. New TestSurveyScaffoldIncumbentProbe extracts the function from SKILL.md and executes it over a committed testdata/scaffold/ fixture pair (workflow-on-disk → spacedock; survey-self-only → nothing) — outcome from on-disk state, not a SKILL.md grep. Live drive over this repo (a genuine spacedock repo): probe now names spacedock via docs/dev/.spacedock-state. 20/20 survey query+probe subtests; 1255/16 full go test; go vet clean. Existing R1-R6 behavior + KEEP-SIGNAL unchanged. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…e report body The render leaned on the bare archetype label "knowledge-work" (HOW YOU WORK line, WORKSTREAMS mode column) — too vague. Now NAME THE SPECIFIC TYPE, synthesized from the workstream names + areas (+ the user's checked-in skills), so "knowledge-work" never stands alone: - SKILL.md step 4 HOW YOU WORK: the knowledge-work line characterizes the actual kinds (e.g. "People 1-1s & team assessment, … memo intake & filing — a knowledge-work loop run mostly autonomously"), not a bare "A knowledge-work loop". - SKILL.md step 4 WORKSTREAMS mode column: qualify the bare label with the track's specific type (e.g. `knowledge-work · 1-1s & team assessment`). The mode-classification VALUE stays `knowledge-work` (the underlying class); only the user-facing characterization names the type. - synthesis-guidance: new "Name the SPECIFIC TYPE of knowledge work" bullet. No sub-type classifier — the type is synthesized from the workstream/area signal the survey already has. Seed a SECOND distinct knowledge-work track (client-1on1s: people 1-1s & assessment) beside notes-ops so the render demonstrably names ≥2 DISTINCT types; mode-classification asserts both → knowledge-work; scoping 15→17. Proven by a live drive over the two-track fixture: HOW YOU WORK + WORKSTREAMS name distinct specific types per track, knowledge-work never standing alone. R1-R6 + cycle-1 (offer reframe, AC-7 scaffold probe) intact. 20/20 query+probe subtests; 1255/16 full go test; go vet clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
clkao
added a commit
that referenced
this pull request
Jun 13, 2026
clkao
added a commit
that referenced
this pull request
Jun 13, 2026
clkao
added a commit
that referenced
this pull request
Jun 13, 2026
clkao
added a commit
that referenced
this pull request
Jun 13, 2026
clkao
added a commit
that referenced
this pull request
Jun 13, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Redesign the
spacedock:surveystep-4 report to lead with plain value and concrete numbers, folding eight user-feedback seeds into one coherent output.What changed
mechanical→manual; add a knowledge-work archetypeEvidence
go test ./...green (15 pkgs)5wv