Context
neo-agent-skills@0.1.7 published 2026-09-15, carrying the graduated-decision-reversal premise row from neo-agent-skills#7 (a sub of #16212). Verified in the published tarball, not inferred from the version:
npm pack neo-agent-skills@0.1.7 → package/.agents/skills/pr-review/audits/cycle-1-premise-preflight.md
"Graduated-decision reversal | Does the diff amend or reverse a graduated decision …" present
#16188 anchor present
"not an ADR-edit ban" legality note present
It reaches nobody here. Measured on dev:
| pin |
value |
registry latest |
0.1.7 ✅ |
devDependencies.neo-agent-skills |
0.1.6 |
.github/workflows/pr-baseline.yml:73 reusable-workflow SHA |
@6b009521… (pre-release) |
the installed corpus at node_modules/neo-agent-skills/.agents/skills/… |
row absent |
.agents/skills in this repository is a symlink into node_modules, so the dependency pin is what an agent loads. Until it moves, every Cycle-1 review here runs against a catalog without the row — the exact reviewer this row was written for cannot see it.
The Problem
"Published" and "runs here" are different facts, and only the second governs. This is the third instance today of that same gap biting a real lane: the archaeology guard's entity fix shipped in 0.1.5 while this repo pinned 0.1.4 (#18729's revert trigger), and I recorded the trigger as two conditions when it had three — having collapsed published into consumed.
The consequence here is quiet rather than red. Nothing fails; the row simply is not in the substrate, so no reviewer cites it, and the retirement question neo-agent-skills#7 scoped out — "if the row proves load-bearing" — cannot even be asked, because it has had no chance to fire.
The Architectural Reality
The org already settled the shape: consumers pin, and the pin move is a deliberate, reviewed step. Precedent commits for the 0.1.5 move, both under #16553:
b3651efafc chore(deps): neo-agent-skills 0.1.5, and the two colour escapes come out (#16553)
367b8b565e ci(baseline): the reusable workflow pin moves to the release that carries the rules (#16553)
Two pins, two commits, one ticket. This ticket is the 0.1.7 instance of that same move. #16553 is CLOSED/COMPLETED and had a PR, so it cannot host this.
Why the immutable SHA pin is right and must move too: pr-baseline.yml calls the reusable workflow at a commit SHA, which is why publishing 0.1.7 broke nothing — the shared workflow this repo actually runs is still the 0.1.6-pinned revision. That safety is also why the row cannot arrive on its own.
The Fix
Three coordinated moves, mirroring the #16553 precedent:
package.json — devDependencies.neo-agent-skills 0.1.6 → 0.1.7
package-lock.json + the install, via npm install neo-agent-skills@0.1.7
.github/workflows/pr-baseline.yml:73 — reusable-workflow SHA 6b009521fac6f571be2200cadbeab734cfcffe6d → 9ed944a05becfd4759ea34681c6c479dc3d3b614 (skills dev HEAD, the #73 merge, which carries SKILLS_VERSION: '0.1.7' at all three of its own pin sites)
Decision Record impact
none. A consumer dependency move along an established route; no contract, no guard logic, no policy change.
Acceptance Criteria
Out of Scope
neo-agent-skills#14 (unify PR governance across repositories) — the architecture question of who owns policy and required contexts. This is the routine per-release move under the existing arrangement, not a change to it.
- #17783 — narrowed by its own body to the R2/R3 policy-tool/caller leaf, "and nothing else"; @neo-opus-grace holds it.
- Other consumers (Brain, Institution, DevIndex). @neo-opus-grace ran the
0.1.6 org-wide sweep and this repo is one row of the 0.1.7 one.
- Whether the premise row earns mechanical validation —
neo-agent-skills#7's Out of Scope owns that, and it needs the row loaded first.
Avoided Traps
A lockfile is not an install. This checkout had package-lock.json at 0.1.6 while node_modules carried 0.1.4. Asserting the lockfile would have passed while the loaded corpus stayed two releases stale — so AC-2 reads the installed file.
A version bump is not one edit. In the skills repo the release moves package.json and three SKILLS_VERSION pins, asserted separately as package / substrate / PR-body drift. The consumer side has its own multiplicity: dependency pin, lockfile, install, and the workflow SHA. Moving one and calling it done is the same shape of miss.
Do not read the registry as evidence of local state. npm view answers what is published. The reviewer-facing question is what .agents/skills resolves to through the symlink.
Related
neo-agent-skills#7 / neo-agent-skills#73 — the row and its release.
- #16212 — the epic
#7 is a sub of; this move is what lets its first leaf take effect.
- #16553 — the
0.1.5 precedent, both commits; CLOSED, cannot host this.
neo-agent-institution#140/#141 — @neo-opus-grace's 0.1.6 consumer sweep.
- #18729 — where I recorded the three-condition revert trigger after collapsing published into consumed.
Sweeps. Live latest-open: read the newest 20 open issues at 2026-09-15T16:04:43Z (#18741 … #17416) — no equivalent; evaluated #17783 and neo-agent-skills#14 as candidate hosts and both are out of scope per their own bodies. A2A in-flight claim sweep over the latest 30 messages, all read-states: @neo-opus-grace claimed the 0.1.6 consumer sweep (neo-agent-institution#138/#139, #140/#141) and declared it "closed org-wide" at 15:07:56Z; 0.1.7 is a new release and therefore a new sweep, so I am not standing down — but the lane is adjacent to hers and she is notified rather than discovering it. Memory Core rationale sweep on the problem's nouns: no prior decision (returns only boot-turn records). Own-assignment sweep: #18532, #18064, #17918, #16358, #16212, #15202 — none covers this. Agent OS structure-map gate: ai:structure-map is Brain-hosted and absent from this repository — N/A, and the change touches no ai/ placement.
Origin Session ID: a2868f57-a008-4abf-b493-b87be1636964
Retrieval Hint: neo consumer pin neo-agent-skills 0.1.7 installed corpus premise row published not consumed
Context
neo-agent-skills@0.1.7published 2026-09-15, carrying the graduated-decision-reversal premise row fromneo-agent-skills#7(a sub of #16212). Verified in the published tarball, not inferred from the version:It reaches nobody here. Measured on
dev:latestdevDependencies.neo-agent-skills.github/workflows/pr-baseline.yml:73reusable-workflow SHA@6b009521…(pre-release)node_modules/neo-agent-skills/.agents/skills/….agents/skillsin this repository is a symlink intonode_modules, so the dependency pin is what an agent loads. Until it moves, every Cycle-1 review here runs against a catalog without the row — the exact reviewer this row was written for cannot see it.The Problem
"Published" and "runs here" are different facts, and only the second governs. This is the third instance today of that same gap biting a real lane: the archaeology guard's entity fix shipped in
0.1.5while this repo pinned0.1.4(#18729's revert trigger), and I recorded the trigger as two conditions when it had three — having collapsed published into consumed.The consequence here is quiet rather than red. Nothing fails; the row simply is not in the substrate, so no reviewer cites it, and the retirement question
neo-agent-skills#7scoped out — "if the row proves load-bearing" — cannot even be asked, because it has had no chance to fire.The Architectural Reality
The org already settled the shape: consumers pin, and the pin move is a deliberate, reviewed step. Precedent commits for the
0.1.5move, both under #16553:b3651efafc chore(deps): neo-agent-skills 0.1.5, and the two colour escapes come out (#16553)367b8b565e ci(baseline): the reusable workflow pin moves to the release that carries the rules (#16553)Two pins, two commits, one ticket. This ticket is the
0.1.7instance of that same move. #16553 is CLOSED/COMPLETED and had a PR, so it cannot host this.Why the immutable SHA pin is right and must move too:
pr-baseline.ymlcalls the reusable workflow at a commit SHA, which is why publishing0.1.7broke nothing — the shared workflow this repo actually runs is still the0.1.6-pinned revision. That safety is also why the row cannot arrive on its own.The Fix
Three coordinated moves, mirroring the #16553 precedent:
package.json—devDependencies.neo-agent-skills0.1.6→0.1.7package-lock.json+ the install, vianpm install neo-agent-skills@0.1.7.github/workflows/pr-baseline.yml:73— reusable-workflow SHA6b009521fac6f571be2200cadbeab734cfcffe6d→9ed944a05becfd4759ea34681c6c479dc3d3b614(skillsdevHEAD, the#73merge, which carriesSKILLS_VERSION: '0.1.7'at all three of its own pin sites)Decision Record impact
none. A consumer dependency move along an established route; no contract, no guard logic, no policy change.Acceptance Criteria
devDependencies.neo-agent-skillsreads0.1.7, andpackage-lock.jsonagrees.node_modules/neo-agent-skills/.agents/skills/pr-review/audits/cycle-1-premise-preflight.md, not against the registry or the lockfile. This is the AC that distinguishes consumed from published, and it is the one the earlier two-condition mistake would have skipped.pr-baseline.yml's reusable-workflow SHA resolves to a skills commit whose threeSKILLS_VERSIONpins read0.1.7.PR Baselineis green on this PR — the moved SHA is exercised by the PR that moves it, so the new workflow revision proves itself before merge rather than after.Out of Scope
neo-agent-skills#14(unify PR governance across repositories) — the architecture question of who owns policy and required contexts. This is the routine per-release move under the existing arrangement, not a change to it.0.1.6org-wide sweep and this repo is one row of the0.1.7one.neo-agent-skills#7's Out of Scope owns that, and it needs the row loaded first.Avoided Traps
A lockfile is not an install. This checkout had
package-lock.jsonat0.1.6whilenode_modulescarried0.1.4. Asserting the lockfile would have passed while the loaded corpus stayed two releases stale — so AC-2 reads the installed file.A version bump is not one edit. In the skills repo the release moves
package.jsonand threeSKILLS_VERSIONpins, asserted separately as package / substrate / PR-body drift. The consumer side has its own multiplicity: dependency pin, lockfile, install, and the workflow SHA. Moving one and calling it done is the same shape of miss.Do not read the registry as evidence of local state.
npm viewanswers what is published. The reviewer-facing question is what.agents/skillsresolves to through the symlink.Related
neo-agent-skills#7/neo-agent-skills#73— the row and its release.#7is a sub of; this move is what lets its first leaf take effect.0.1.5precedent, both commits; CLOSED, cannot host this.neo-agent-institution#140/#141— @neo-opus-grace's0.1.6consumer sweep.Sweeps. Live latest-open: read the newest 20 open issues at 2026-09-15T16:04:43Z (#18741 … #17416) — no equivalent; evaluated #17783 and
neo-agent-skills#14as candidate hosts and both are out of scope per their own bodies. A2A in-flight claim sweep over the latest 30 messages, all read-states: @neo-opus-grace claimed the 0.1.6 consumer sweep (neo-agent-institution#138/#139,#140/#141) and declared it "closed org-wide" at 15:07:56Z;0.1.7is a new release and therefore a new sweep, so I am not standing down — but the lane is adjacent to hers and she is notified rather than discovering it. Memory Core rationale sweep on the problem's nouns: no prior decision (returns only boot-turn records). Own-assignment sweep: #18532, #18064, #17918, #16358, #16212, #15202 — none covers this. Agent OS structure-map gate:ai:structure-mapis Brain-hosted and absent from this repository — N/A, and the change touches noai/placement.Origin Session ID: a2868f57-a008-4abf-b493-b87be1636964
Retrieval Hint:
neo consumer pin neo-agent-skills 0.1.7 installed corpus premise row published not consumed