topic-docs: dispose of the 17 contract-slice directories on main — prune 8, graduate 3, register 6

[kyle-sexton]

This was generated by AI, delegated directly by the repository operator to resolve #1417's blocker.

Companion to #1417. That issue wires the docs/topics/** prune gate; this one disposes of the 17
contract-slice directories currently on main. The two are independent — #1417's decision
includes a pure-deletion carve-out, so the gate never blocks a prune and the tickets may land in
either order.

Full evidence and rationale: the decision comment on #1417.

Why the 17 are not one population

Reading all 17 slices, their git history, and their linked tracker issues splits them three ways.
A mechanical bulk delete would lose durable information; a per-slice disposition does not.

Phase A — mechanical prune (8 slices, no triage needed)

Delete these outright. All phases [DONE], shipped artifacts present in the tree, no un-routed
follow-ups, zero inbound markdown links:

• docs/topics/plugin-fleet-sync-skill/ — 6 phases [DONE] 2026-07-17; plugins/claude-ops/skills/plugins/ ships
• docs/topics/underspecification/ — self-declares "COMPLETED (recorded 2026-07-21) … no further work"
• docs/topics/handoff-bg-cutover/ — every step shipped; plugins/session-flow/skills/continue-in-background/ exists
• docs/topics/plugin-organization/ — ## Plan is literally "(unfilled)"; grammar + vocabulary graduated to docs/CATALOG-TAXONOMY.md
• docs/topics/github-plugin-candidates/ — 7 phases [DONE]; plugins/github/skills/{audit,advise,setup} ship. Also actively stale: cites the retired consumer-config-layering path in ~6 places
• docs/topics/dometrain-mcp/ — 7 phases [DONE], plugin ships. Two unresolved empirical questions (bad-key 401 behavior; duplicate plugin.json name behavior) are recorded only here — route to the tracker or drop them deliberately before deleting
• docs/topics/babysit-prs-migration/ — stale [DOING]/[TODO] marks from 07-18 for an engine that already shipped (babysit-prs + babysit-loop both exist)
• docs/topics/boris-video-absorption/ — 8 phases [DONE], "Open questions: None blocking". Its deferred-with-trigger items live only in a machine-local digest slice; decide explicitly whether any are worth a tracker issue first

Phase B — cheap graduation, then prune (3 slices)

Each needs one human-authored paragraph promoted somewhere durable before the directory goes. Do not
fold these into Phase A's delete.

• docs/topics/commit-convention-well-known-path/ — the decision and its 3-rung precedence already
  graduated to docs/conventions/commit-convention/README.md:80-113. Not graduated: the rejected
  alternatives, and a tracked concern that "the operator does not recall ratifying
  consumer-config-layering … audit which seams were human-ratified vs agent-accreted" — that one is
  a tracker issue, not a doc paragraph. Also fix the dangling backtick ref at
  docs/conventions/commit-convention/README.md:104
• docs/topics/plugin-audit-port/ — A1–A5 / B1–B6 [DONE], both plugins ship. B7 "operator cutover
  (post-merge, HITL)" is [TODO] and needs a home; four open questions (zone bands, stable-shim,
  zones.json org layer) are USER-RESERVED and must route to the tracker, not be deleted. Also fix
  the dangling ref at docs/adr/0005-…:326
• docs/topics/fable-field-guide-audit/ — remediations shipped via feat(playbooks): adopt the fable field guide audit remediations into fable-5 #1261 and
  plugins/playbooks/skills/fable-5/. Not graduated: the "dropped"/"hold" rulings and
  codex-review.md §4's standing objection. Separately, this slice carries raw-capture.txt
  and source-article.md — a raw command capture and verbatim third-party article text, both
  forbidden by the convention's redaction bar (docs/conventions/topic-docs/README.md:315-320).
  That is a live violation regardless of this ticket's outcome; handle it here since the directory
  is being touched anyway

Phase C — register hygiene (6 live programs, no deletion)

These stay. They are multi-branch program slices, not violations — see #1417's decision. This phase
only ensures each one is legitimately registrable:

Slug	State	Action
autonomy-ignition	#778 OPEN, commit today (#1363)	register against #778
fresh-eyes-checkpoint-audit	#304 OPEN, 0 of 4 phases executed	register against #304
context-engineering-claude-5	"in progress", 5 phases [TODO]	needs an open tracker issue — file one
ladder-climb-roadmap	1 of 5 phases [DONE]	needs an open tracker issue — file one, or register against #778 if genuinely the same program
loop-engineering-codification	4 un-routed follow-ups, 2 open interview branches	needs an open tracker issue — file one; route the four follow-ups
ai-adoption-ladder	T1–T7 graduated to plugins/autonomy/reference/; 10 RESEARCH-*.md and native-vs-hook-telemetry-audit.md did not	needs an open tracker issue. The telemetry audit is a version-pinned decision record with revisit triggers — it is ADR-shaped and is the strongest ADR candidate in the whole corpus

Hard constraint

Do not delete docs/topics/loop-engineering-codification/. Its PLAN holds the sole record of a
USER-RESERVED, still-open question — "whether contract slices should persist on the default branch
at all" — which is the premise of this entire ticket and which #1417's decision explicitly declined
to answer. Deleting it destroys the record of its own premise.

Verification

• git ls-tree -d --name-only origin/main docs/topics/ returns exactly the 6 registered program
  slugs when all phases are complete
• No new dangling references: git grep -nE '\]\([^)]*docs/topics/[^)]*\)' -- . ':!docs/topics/'
  stays empty (it is empty today — there are zero inbound markdown links, so no link-checker risk)
• Every USER-RESERVED question and un-routed follow-up named above exists as a tracker issue before
  its slice is deleted
• If topic-docs: the contract-slice prune gate is specified but never wired — 17 slices on main #1417 has already landed, each phase's PR passes topic-slice-prune-gate — pure-deletion diffs
  pass by the carve-out

Suggested PR shape

One PR per phase. Phase A is a single mechanical deletion PR and is CI-cheap (docs/topics/ is on
scripts/docs-only-paths.txt:43, so the heavy lanes report evaluated-and-not-applicable). Phases B
and C carry judgment and should not ride along with it.

[kyle-sexton]

This was generated by AI, delegated directly by the repository operator to resolve #1417's blocker.

Two clarifications, mirroring the ones just posted on #1417.

Phase C does not block #1417

#1417 may seed its register with all six program slugs immediately; the three that lack a dedicated
tracker issue (context-engineering-claude-5, ladder-climb-roadmap,
loop-engineering-codification) reference this issue as their live tracker reference. Phase C
then replaces those references with the real issue numbers as it files them. Neither ticket waits on
the other.

Phases A and B must remove register entries in the same PR

Per #1417's precedence rule, register hygiene is evaluated against the merge result and wins over
the deletion carve-out: a PR that prunes a registered slice must remove that slug from
scripts/topic-slice-programs.txt in the same PR, or the gate fails naming the stale entry.

This does not affect Phase A or B as scoped today — none of the eight drained slices or the three
mid-tier slices are registered. It becomes load-bearing later, when a Phase C program closes and its
slice is pruned. Noted here so the eventual prune PR does not trip the gate.

[kyle-sexton]

[kyle-sexton]

This was generated by AI during work-loop execution.

Phase A landed: #1459 (CI green) prunes the 8 mechanical-complete slices —
plugin-fleet-sync-skill, underspecification, handoff-bg-cutover, plugin-organization,
github-plugin-candidates, dometrain-mcp, babysit-prs-migration, boris-video-absorption —
and drops their slugs from scripts/contract-slice-baseline.txt. Per this issue's own "Suggested PR
shape," Phase A rode alone; Phases B and C remain open under this issue.

Two of the eight carried a precondition before deletion. Discharged via new tracker items rather
than letting deletion drop the record:

• dometrain-mcp: two unresolved MCP-host behavior questions (401 rendering, duplicate plugin.json name) need a live test #1456 — dometrain-mcp's two unresolved MCP-host behavior questions (bad-key 401 rendering;
  duplicate plugin.json name behavior)
• boris-video-absorption: three deferred-with-trigger follow-ups need to stay discoverable after slice prune #1457 — boris-video-absorption's three deferred-with-trigger items (verification self-improvement
  loop, bounded build→review→fix loop, Claude Tag enablement), none of whose triggers has fired

Not started: Phase B (graduate-then-prune: commit-convention-well-known-path,
plugin-audit-port, fable-field-guide-audit) and Phase C (register hygiene: the 6 live-program
slugs). Both carry judgment calls this pass didn't make.

Inventory note: docs/topics/ actually held 19 directories at audit time, not 17 — #1400
landed interview-batch-rounds and shadowed-skill-renames 2 minutes before this issue was filed.
Both are already accounted for as pre-existing debt in scripts/contract-slice-baseline.txt and
tracked under #1419; no action taken on them here since this issue's own scope is the named 17.