topic-docs: graduate and prune the 17 contract slices already on main

[kyle-sexton]

main carries 17 contract-slice directories under docs/topics/ that the topic-docs
convention says should have been pruned before their PRs merged. The missing enforcement is #1417;
this issue is the debt those 17 represent.

Why this is separate from #1417

#1417 wires the gate. The gate stops new slices from landing, but it deliberately does not
delete anything — pruning a slice is not a mechanical operation. Per
docs/conventions/topic-docs/README.md, step 4 prunes the slice only after step 3 graduates its
durable outcomes:

architectural decisions and specs through the knowledge-vault seam (default:
history-preserving git mv into docs/adr/ / docs/specs/; remote vault backends resolve
through the same seam), and actionable follow-ups through the work-item tracker seam

So each slice needs a judgement call by someone who knows what it was for: what graduates to an
ADR or spec, what becomes a tracker item, and what is genuinely spent and can be deleted. That is
17 independent judgements, most of them on work this issue's author did not do.

Inventory

71 files, ~1.27 MB.

Slice	Files	Bytes
ai-adoption-ladder	13	203,842
autonomy-ignition	2	28,453
babysit-prs-migration	3	65,654
boris-video-absorption	2	15,153
commit-convention-well-known-path	1	9,623
context-engineering-claude-5	11	368,603
dometrain-mcp	2	60,190
fable-field-guide-audit	22	294,587
fresh-eyes-checkpoint-audit	2	34,009
github-plugin-candidates	2	41,847
handoff-bg-cutover	1	5,565
ladder-climb-roadmap	3	29,673
loop-engineering-codification	1	18,352
plugin-audit-port	2	51,096
plugin-fleet-sync-skill	2	34,884
plugin-organization	1	9,092
underspecification	1	3,294

Four of them are load-bearing for open PRs right now

These must not be deleted out from under in-flight work:

Open PR	Slice it touches
#1318	context-engineering-claude-5
#1252	plugin-audit-port
#1096	fresh-eyes-checkpoint-audit
#1400	adds two new slices (interview-batch-rounds, shadowed-skill-renames)

#1400 is worth calling out: it is an open PR whose stated purpose is to land two more contract
slices. Whatever disposition this issue reaches should cover the slices #1400 adds, or #1400 should
be redirected before it merges.

Suggested shape

One PR per slice, or one PR per owner — not a single 71-file deletion. Each PR:

1. Graduates what is durable (ADR / spec / tracker item).
2. Deletes the slice directory.
3. Removes that slug's line from scripts/contract-slice-baseline.txt.

The baseline is stale-guarded: once a slice is gone, its baseline entry starts failing --check,
so the exemption cannot outlive the debt and this issue cannot be quietly abandoned half-done.

Done when

scripts/contract-slice-baseline.txt contains no slugs, and docs/topics/ is empty on main.

[kyle-sexton]

Inventory correction: 19 slices, not 17.

#1400 merged after this issue was filed (b8554481, 2026-07-25T23:43:47Z), landing two more:

• interview-batch-rounds
• shadowed-skill-renames

Both are in scripts/contract-slice-baseline.txt on #1429 as pre-existing debt rather than
incoming work — by the time the gate lands they are on main, so red-lining unrelated PRs for them
would be wrong. They belong to this burn-down like the other 17.

Worth recording that this happened during the ~4 hours #1429 was open, which is the strongest
available argument for landing the gate before the cleanup: the failure mode is live, not
historical.

The table above is otherwise unchanged. The "done when" condition still holds — this issue closes
when scripts/contract-slice-baseline.txt lists no slugs and docs/topics/ is empty on main.

Also note the four "load-bearing for open PRs" rows have shifted, since #1400 is now merged:

Open PR	Slice	Status
#1252	plugin-audit-port	passes on baseline exemption
#1096	fresh-eyes-checkpoint-audit	passes on baseline exemption
#1318	context-engineering-claude-5	needs a rebase past #1400 before the gate reads it correctly

[kyle-sexton]

This was generated by AI (triage lane — gate-audit claim). Claiming this item for one gate-audit cycle: an independent panel re-classifies whether the needs-human / status: needs-decision gate is still warranted, or whether the item is C2/C3 and liftable to agent-ready. Outcome comment follows in this cycle.

[kyle-sexton]

This was generated by AI (triage lane — gate audit). Independent panel: two fresh-context agents (one fable, one opus), each given only the issue content and the repo's own plugins/autonomy/reference/guardrails/work-classes.md, plus the triage lane's own read. This lane never executes work items, authors fixes, or merges.

Gate audit: GATE UPHELD — C4 structural, irreversible, 19 independent judgments

Panel: 2-of-2 STAYS-GATED, both C4. Lane read concurs. Multiple floors stack here.

The issue states pruning "is not a mechanical operation" and that each slice needs "a judgement call by someone who knows what it was for" — the author's own comment counts 17 independent judgements, most on work this issue's author did not do, corrected upward to 19 after #1400 merged. The work ends in deletion of roughly 1.27 MB across 71+ files, which is the hard-to-reverse floor: graduation decisions determine what knowledge survives as ADRs/specs before the slices are gone.

Sequencing hazard, verified by the panel: three slices are load-bearing for open PRs (#1252 and #1096 pass on baseline exemption; #1318 needs a rebase past #1400). Mis-sequenced deletion breaks in-flight work.

Decision required

For each of the 19 slices: what is durable and graduates to an ADR/spec or tracker item, versus genuinely spent and deletable — and in what order?

RECOMMENDED — one human triage pass recording a per-slice disposition, then agent-executed PRs one slice at a time. The maintainer posts a checklist comment assigning each of the 19 a disposition (graduate / file / delete); agents then execute one PR per slice, each doing the history-preserving git mv into docs/adr/ or docs/specs/, deleting the directory, and removing that slug's line from scripts/contract-slice-baseline.txt so the stale-guard prevents a half-abandoned burn-down. Sequence the three PR-load-bearing slices last. This splits the judgment (human, cheap, one pass) from the execution (agent, mechanical, 19 PRs) and matches the issue's own suggested shape.

Alternative — fully human-driven per-slice PRs by the original slice owners. Slower, but keeps judgment and execution in the same hands, which matters most where the author no longer remembers the intent.

Rejected — a single sweeping 71-file deletion PR; the issue argues against it explicitly and it forfeits per-slice reversibility.

[kyle-sexton]

This was generated by AI (triage lane — claim release). Gate-audit cycle complete for this item; the triage claim posted earlier this cycle is released. No lease is held — another instance or lane may pick this up immediately. The outcome comment above this one carries the classification and either the lift rationale or the decision summary.