@codyswann/lisa 4.55.2 → 4.55.4

package/plugins/src/base/skills/lisa-github-build-intake/SKILL.md

@@ -6,6 +6,20 @@ allowed-tools: ["Skill", "Bash"]
 
 # GitHub Build Intake: $ARGUMENTS
 
+## Human-gate release authorization
+
+A release requires a trusted human author, not just matching comment text. Follow
+`ready-role-filing` — **Human-gate release authorization**: preserve tracker-supplied comment
+author IDs and bot metadata, resolve `trustedHumanActorIds` only from an explicit user instruction
+or existing human-authored trusted project policy, and pass it with structured `comments` to every
+hold classifier, reconciliation, normalization and release planner. Never derive trust from the
+comment body, a display name, the actor's own assertion, or an automation posting on its own behalf.
+Missing policy, missing/unreadable author identity, raw body strings, untrusted actors and known bots
+cannot discharge a hold. Keep the item held and report the missing authorization; do not silently
+replace these inputs with an empty history or an inferred allowlist. Authorized matching releases
+continue through the existing path and never override an independently declared caller hold.
+
+
 `$ARGUMENTS` is one of:
 
 1. A GitHub `org/repo` token (e.g., `acme/frontend-v2`).
@@ -215,14 +229,21 @@ nothing survives, exit on the denominator-stated summary from `summarizeDryLane`
 A blocker is a **claim with a timestamp, not a fact** — it goes stale the moment its condition comes
 true, and nothing re-read one before this phase. For each `$BLOCKED` candidate:
 
-1. **Human gate first, and it is absolute.** An issue carrying the configured human-needed label
-   (`github.labels.build.human_needed`, default `human-needed`) or a `[lisa-human-gate]` marker in
-   its body is **never** auto-selected, whatever any probe says.
+1. **Read release comments before applying the human gate.** Pass all item comments, labels,
+   body, `trustedHumanActorIds` and configured `humanNeededLabel` to `classifyReadyCandidate(...)`. For a discharged hold,
+   call `planHumanGateRelease({ labels, body, comments, trustedHumanActorIds, humanNeededLabel, readyLabel, lifecycleLabels, alreadyNotified })`
+   and retain its plan. Apply human-marker cleanup only as planned; defer any ready-role restoration
+   until steps 2–5 have cleared the remaining blockers (map the role to the vendor's state or label).
+   A still-active hold is never auto-selected, whatever any probe says. An absent or unreadable
+   comment history cannot discharge a hold. A release removes only the human hold: continue the
+   remaining blocker checks below before selecting the item.
+
 2. **Extract the stated discharge condition**, then **probe it**. Machine-testable conditions — a
    version on trunk, a published package, a CI run history, an advisory's patched status — rot
    fastest and are cheapest to check. A human decision is not machine-testable; leave it.
-3. **Classify with `classifyPreWorkCandidate(...)`** from `scripts/intake-blocker-reprobe.mjs`. A
-   discharge with no recorded evidence is not a discharge, and neither is a candidate nothing
+3. **Classify with `classifyPreWorkCandidate(...)`** from `scripts/intake-blocker-reprobe.mjs`.
+   Include the structured `comments` and `trustedHumanActorIds` from step 1.
+   A discharge with no recorded evidence is not a discharge, and neither is a candidate nothing
    probed this cycle — the helper refuses both.
 4. **Record the result on the issue either way** via `formatReprobeNote(...)` as a comment, so the
    next cycle reads the answer rather than re-deriving it. Keep it idempotent.
@@ -317,7 +338,7 @@ other gate's verdict — however conclusive — may promote an item a person par
    drift, and a drifted gate fails *silently*, by quietly ceasing to match. Do **not** re-implement
    the test here, and do **not** key it on `reason=`: markers in the wild carry no `reason=` key at
    all and sit anywhere in the body, so a structured parse would miss them while appearing to work
-   on every item that happens to have one. **Pass the item's `comments` alongside its labels and
+   on every item that happens to have one. **Pass the item's structured `comments` and `trustedHumanActorIds` alongside its labels and
    body.** A hold is ended by a release recorded in a comment, so a reader handed no comments cannot
    see the discharge — it goes on holding an item whose question was answered weeks ago, which is
    the defect this gate carried from the day it was written (CodySwannGT/lisa#3852). Omitting them
@@ -328,16 +349,17 @@ other gate's verdict — however conclusive — may promote an item a person par
    and re-rejected every cycle forever and seen by nothing — `lisa-repair-intake` sweeps items that
    are **not** in the ready role and excludes gated ones outright, so a ready-and-gated item falls
    outside its filter twice over. Call
-   `planHumanGateReconciliation({ labels, body, humanNeededLabel, readyLabel, alreadyNotified })`
+   `planHumanGateReconciliation({ labels, body, comments, trustedHumanActorIds, humanNeededLabel, readyLabel, alreadyNotified })`
    and apply exactly the actions it returns: remove `$READY`, add the configured human-needed
    marker, and post `formatHumanGateNote()` once. The planner is idempotent by state, so an item
    already out of the lane and already marked yields no second mutation and no second comment. This
    is the same repair the leaf-only gate already performs for a ready item that must not be
    dispatched.
 4. **On `claimable: true` for an item that still carries a hold, RELEASE it — do not just proceed.**
-   The hold left durable state behind: the item is out of the queue and flagged as needing a person,
-   and answering the question does not undo either. Call
-   `planHumanGateRelease({ labels, body, comments, humanNeededLabel, readyLabel, lifecycleLabels, alreadyNotified })`
+   This ready-lane candidate can retain a historical body marker and a human-needed label.
+   This step reconciles only candidates still in the ready lane; released holds outside it are
+   recovered by `lisa-repair-intake` step 2b (or Phase 2.5 when included in this cycle). Call
+   `planHumanGateRelease({ labels, body, comments, trustedHumanActorIds, humanNeededLabel, readyLabel, lifecycleLabels, alreadyNotified })`
    and apply exactly the actions it returns: remove the configured human-needed marker, add the
    configured ready role back, and post `formatHumanGateReleaseNote()` once. It is the exact inverse
    of step 3's planner and it refuses in both directions — an item still held plans nothing, and an
@@ -369,11 +391,12 @@ GitHub Issues live in one repo by definition, so the scanned repo's issues are u
 1. **Resolve the current repo** per `config-resolution` "Repo scoping" (`.repo` → `.github.repo` → `git remote get-url origin` basename). If unresolvable, stop and report.
 2. **Cheap path first.** Prefer candidates already carrying the `repo:<current>` label. Keep the Phase 2 scan broad so unlabeled issues are still seen, determined, and stamped.
 3. **Per candidate, apply the repo-scope decision (`repo-scope-split`):**
-   - Carries `repo:<other>` → **skip** (leave it `ready` for that repo's own intake); next candidate.
-   - **Unlabeled** → determine the target repo(s) from the issue + code surfaces, then **stamp** `repo:<name>` via `gh issue edit <n> --add-label "repo:<name>"` (create the label lazily) so later cycles filter cheaply; re-apply with the now-known repo. (An issue whose work is entirely in the scanned repo is simply labeled `repo:<current>`.)
-   - **Container visibility is allowed.** A multi-repo Epic / Story / Spike may legitimately carry multiple `repo:<name>` labels for operator visibility. Do not split or claim it here; leave the repo markers intact and fall through to the leaf-only gate, which repairs the stale build-ready label instead of dispatching the container.
-   - **Multi-repo leaf → split, never claim.** Run the `repo-scope-split` work-time procedure into single-repo siblings, each created **build-ready** (`build_ready: true`) and stamped with its own `repo:<name>`; the current repo's sibling becomes a normal candidate.
-   - **Single-repo leaf for the current repo** → fall through to 3a (leaf-only gate) and 3b (claim).
+   - **Count distinct repository markers first**: collect all `repo:<name>` labels and, on JIRA, recognized repository components; deduplicate by repository. A container may carry multiple `repo:<name>` labels. Do not split or claim it here; send it to the leaf-only gate.
+   - **Multi-repo leaf → split, never claim.** More than one repository takes the work-time split before any wrong-repository skip. Each sibling is **build-ready** (`build_ready: true`) and stamped with its own `repo:<name>`; the current repo's sibling becomes a normal candidate.
+   - **Exactly one other repository** (`repo:<other>`) → **skip**, leaving the item ready for that repo's intake.
+   - **No repository marker** → determine target repo(s) from the ticket and code, stamp `repo:<name>` through the vendor access layer, and re-apply from the count.
+   - **Single-repo leaf for the current repo** → fall through to 3a and 3b.
+
 4. Continue until a claimable current-repo leaf is found (claim it; one per cycle) or the candidate set is exhausted — exit cleanly on the denominator-stated summary, naming the current repo alongside the swept lanes.
 
 #### 3a. Leaf-only claim gate (repair containers)

package/plugins/src/base/skills/lisa-github-write-issue/SKILL.md

@@ -6,6 +6,20 @@ allowed-tools: ["Bash", "Skill", "Read"]
 
 # Write GitHub Issue: $ARGUMENTS
 
+## Human-gate release authorization
+
+A release requires a trusted human author, not just matching comment text. Follow
+`ready-role-filing` — **Human-gate release authorization**: preserve tracker-supplied comment
+author IDs and bot metadata, resolve `trustedHumanActorIds` only from an explicit user instruction
+or existing human-authored trusted project policy, and pass it with structured `comments` to every
+hold classifier, reconciliation, normalization and release planner. Never derive trust from the
+comment body, a display name, the actor's own assertion, or an automation posting on its own behalf.
+Missing policy, missing/unreadable author identity, raw body strings, untrusted actors and known bots
+cannot discharge a hold. Keep the item held and report the missing authorization; do not silently
+replace these inputs with an empty history or an inferred allowlist. Authorized matching releases
+continue through the existing path and never override an independently declared caller hold.
+
+
 Create or update a GitHub Issue with all required relationships, metadata, and quality gates. Every section below is mandatory. Thin issues are rejected.
 
 This skill is the GitHub counterpart of `lisa-jira-write-ticket`. The two skills share the same gates, description structure, acceptance-criteria format, and verification flow. The data-model translation is documented under "GitHub Issues data model" below — keep that table in sync with `lisa-github-validate-issue` so what one skill writes the other accepts.
@@ -323,7 +337,7 @@ If the label does not exist in the tracker, create it, or record that it could n
 proceed — the marker still holds. Never file the label *instead of* the marker.
 
 **Write a `reason=` the release can name.** The hold's reason is not decoration: a hold ends when a
-`[lisa-human-gate-release]` comment repeating that same `reason=` is recorded on the item, and the
+`[lisa-human-gate-release]` comment repeating that same `reason=` is recorded on the item by an authorized human, and the
 next intake sweep then takes the marker label off and puts the item back in the build-ready role on
 its own. Matching is per-reason so that a hold declared *after* an earlier release is not born
 discharged. A keyless hold is legal and is discharged by a keyless release; a hold whose reason is a

package/plugins/src/base/skills/lisa-jira-build-intake/SKILL.md

@@ -6,6 +6,20 @@ allowed-tools: ["Skill", "Bash"]
 
 # JIRA Build Intake: $ARGUMENTS
 
+## Human-gate release authorization
+
+A release requires a trusted human author, not just matching comment text. Follow
+`ready-role-filing` — **Human-gate release authorization**: preserve tracker-supplied comment
+author IDs and bot metadata, resolve `trustedHumanActorIds` only from an explicit user instruction
+or existing human-authored trusted project policy, and pass it with structured `comments` to every
+hold classifier, reconciliation, normalization and release planner. Never derive trust from the
+comment body, a display name, the actor's own assertion, or an automation posting on its own behalf.
+Missing policy, missing/unreadable author identity, raw body strings, untrusted actors and known bots
+cannot discharge a hold. Keep the item held and report the missing authorization; do not silently
+replace these inputs with an empty history or an inferred allowlist. Authorized matching releases
+continue through the existing path and never override an independently declared caller hold.
+
+
 All Atlassian operations in this skill go through `lisa-atlassian-access`. Do not call MCP tools or `acli` directly.
 
 `$ARGUMENTS` is one of:
@@ -189,14 +203,21 @@ swept (see `automation-runbook-contract`).
 A blocker is a **claim with a timestamp, not a fact** — it goes stale the moment its condition comes
 true, and nothing re-read one before this phase. For each pre-work candidate outside `$READY`:
 
-1. **Human gate first, and it is absolute.** A ticket carrying the configured human-needed label
-   (`jira.labels.human_needed`, default `Human Needed`) or a `[lisa-human-gate]` marker in its
-   description is **never** auto-selected, whatever any probe says.
+1. **Read release comments before applying the human gate.** Pass all item comments, labels,
+   body, `trustedHumanActorIds` and configured `humanNeededLabel` to `classifyReadyCandidate(...)`. For a discharged hold,
+   call `planHumanGateRelease({ labels, body, comments, trustedHumanActorIds, humanNeededLabel, readyLabel, lifecycleLabels, alreadyNotified })`
+   and retain its plan. Apply human-marker cleanup only as planned; defer any ready-role restoration
+   until steps 2–5 have cleared the remaining blockers (map the role to the vendor's state or label).
+   A still-active hold is never auto-selected, whatever any probe says. An absent or unreadable
+   comment history cannot discharge a hold. A release removes only the human hold: continue the
+   remaining blocker checks below before selecting the item.
+
 2. **Extract the stated discharge condition**, then **probe it**. Machine-testable conditions — a
    version on trunk, a published package, a CI run history, an advisory's patched status — rot
    fastest and are cheapest to check. A human decision is not machine-testable; leave it.
 3. **Classify with `classifyPreWorkCandidate(...)`** from
-   `scripts/intake-blocker-reprobe.mjs`. A discharge with no recorded evidence is not a discharge,
+   `scripts/intake-blocker-reprobe.mjs`, including the structured `comments` and
+   `trustedHumanActorIds` from step 1. A discharge with no recorded evidence is not a discharge,
    and neither is a candidate nothing probed this cycle — the helper refuses both.
 4. **Record the result on the ticket either way** via `formatReprobeNote(...)` as a comment, so the
    next cycle reads the answer rather than re-deriving it. Keep it idempotent.
@@ -222,7 +243,7 @@ other gate's verdict — however conclusive — may promote an item a person par
    drift, and a drifted gate fails *silently*, by quietly ceasing to match. Do **not** re-implement
    the test here, and do **not** key it on `reason=`: markers in the wild carry no `reason=` key at
    all and sit anywhere in the body, so a structured parse would miss them while appearing to work
-   on every item that happens to have one. **Pass the item's `comments` alongside its labels and
+   on every item that happens to have one. **Pass the item's structured `comments` and `trustedHumanActorIds` alongside its labels and
    body.** A hold is ended by a release recorded in a comment, so a reader handed no comments cannot
    see the discharge — it goes on holding an item whose question was answered weeks ago, which is
    the defect this gate carried from the day it was written (CodySwannGT/lisa#3852). Omitting them
@@ -233,16 +254,17 @@ other gate's verdict — however conclusive — may promote an item a person par
    and re-rejected every cycle forever and seen by nothing — `lisa-repair-intake` sweeps items that
    are **not** in the ready role and excludes gated ones outright, so a ready-and-gated item falls
    outside its filter twice over. Call
-   `planHumanGateReconciliation({ labels, body, humanNeededLabel, readyLabel, alreadyNotified })`
+   `planHumanGateReconciliation({ labels, body, comments, trustedHumanActorIds, humanNeededLabel, readyLabel, alreadyNotified })`
    and apply exactly the actions it returns: remove `$READY`, add the configured human-needed
    marker, and post `formatHumanGateNote()` once. The planner is idempotent by state, so an item
    already out of the lane and already marked yields no second mutation and no second comment. This
    is the same repair the leaf-only gate already performs for a ready item that must not be
    dispatched.
 4. **On `claimable: true` for an item that still carries a hold, RELEASE it — do not just proceed.**
-   The hold left durable state behind: the item is out of the queue and flagged as needing a person,
-   and answering the question does not undo either. Call
-   `planHumanGateRelease({ labels, body, comments, humanNeededLabel, readyLabel, lifecycleLabels, alreadyNotified })`
+   This ready-lane candidate can retain a historical body marker and a human-needed label.
+   This step reconciles only candidates still in the ready lane; released holds outside it are
+   recovered by `lisa-repair-intake` step 2b (or Phase 2.5 when included in this cycle). Call
+   `planHumanGateRelease({ labels, body, comments, trustedHumanActorIds, humanNeededLabel, readyLabel, lifecycleLabels, alreadyNotified })`
    and apply exactly the actions it returns: remove the configured human-needed marker, add the
    configured ready role back, and post `formatHumanGateReleaseNote()` once. It is the exact inverse
    of step 3's planner and it refuses in both directions — an item still held plans nothing, and an
@@ -274,10 +296,12 @@ A JIRA project can oversee multiple repos (`frontend` / `backend` / `infrastruct
 1. **Resolve the current repo** per `config-resolution` "Repo scoping" (`.repo` → `.github.repo` → `git remote get-url origin` basename). If unresolvable, stop and report — do not claim tickets you cannot scope.
 2. **Cheap path first.** Phase 1's query-time pre-filter has already dropped tickets explicitly stamped for a sibling repo, so the Phase 2 result set is current-repo-labeled + unlabeled tickets. Prefer candidates already carrying `repo:<current>` — a JIRA **label**, or a **component** equal to the repo name (accepted as an alias); the pre-filter is label-only, so a ticket scoped solely by a sibling-repo **component** can still appear and is skipped here. The result set still includes unlabeled tickets so they can be determined and stamped; this gate orders/filters what remains.
 3. **Per candidate, apply the repo-scope decision (`repo-scope-split`):**
-   - Carries `repo:<other>` (label or component) → **skip** (leave it `ready` for that repo's own intake); next candidate.
-   - **Unlabeled** → determine the target repo(s) from the ticket (description, AC, technical approach) confirmed against the code surfaces, then **stamp** `repo:<name>` via `lisa-atlassian-access` `operation: write-ticket` (add the label / set the component) so later cycles filter cheaply; re-apply with the now-known repo.
-   - **Multi-repo leaf → split, never claim.** Run the `repo-scope-split` work-time procedure to break it into single-repo siblings, each created **build-ready** (`build_ready: true`) and stamped with its own `repo:<name>`; the current repo's sibling becomes a normal candidate.
-   - **Single-repo leaf for the current repo** → fall through to 3a (leaf-only gate) and 3b (claim).
+   - **Count distinct repository markers first**: collect all `repo:<name>` labels and, on JIRA, recognized repository components; deduplicate by repository. A container may carry multiple `repo:<name>` labels. Do not split or claim it here; send it to the leaf-only gate.
+   - **Multi-repo leaf → split, never claim.** More than one repository takes the work-time split before any wrong-repository skip. Each sibling is **build-ready** (`build_ready: true`) and stamped with its own `repo:<name>`; the current repo's sibling becomes a normal candidate.
+   - **Exactly one other repository** (`repo:<other>`) → **skip**, leaving the item ready for that repo's intake.
+   - **No repository marker** → determine target repo(s) from the ticket and code, stamp `repo:<name>` through the vendor access layer, and re-apply from the count.
+   - **Single-repo leaf for the current repo** → fall through to 3a and 3b.
+
 4. Continue until a claimable current-repo leaf is found (claim it; one per cycle) or the candidate set is exhausted — exit cleanly on the denominator-stated summary, naming the current repo alongside the swept lanes.
 
 #### 3a. Leaf-only claim gate (skip / safe-block containers)

package/plugins/src/base/skills/lisa-jira-write-ticket/SKILL.md

@@ -6,6 +6,20 @@ allowed-tools: ["Bash", "Skill"]
 
 # Write JIRA Ticket: $ARGUMENTS
 
+## Human-gate release authorization
+
+A release requires a trusted human author, not just matching comment text. Follow
+`ready-role-filing` — **Human-gate release authorization**: preserve tracker-supplied comment
+author IDs and bot metadata, resolve `trustedHumanActorIds` only from an explicit user instruction
+or existing human-authored trusted project policy, and pass it with structured `comments` to every
+hold classifier, reconciliation, normalization and release planner. Never derive trust from the
+comment body, a display name, the actor's own assertion, or an automation posting on its own behalf.
+Missing policy, missing/unreadable author identity, raw body strings, untrusted actors and known bots
+cannot discharge a hold. Keep the item held and report the missing authorization; do not silently
+replace these inputs with an empty history or an inferred allowlist. Authorized matching releases
+continue through the existing path and never override an independently declared caller hold.
+
+
 All Atlassian operations in this skill go through `lisa-atlassian-access`. Do not call MCP tools or `acli` directly.
 
 Create or update a JIRA ticket with all required relationships, metadata, and quality gates. Every section below is mandatory. Thin tickets are rejected.
@@ -308,7 +322,7 @@ If the label does not exist in the tracker, create it, or record that it could n
 proceed — the marker still holds. Never file the label *instead of* the marker.
 
 **Write a `reason=` the release can name.** The hold's reason is not decoration: a hold ends when a
-`[lisa-human-gate-release]` comment repeating that same `reason=` is recorded on the item, and the
+`[lisa-human-gate-release]` comment repeating that same `reason=` is recorded on the item by an authorized human, and the
 next intake sweep then takes the marker label off and puts the item back in the build-ready role on
 its own. Matching is per-reason so that a hold declared *after* an earlier release is not born
 discharged. A keyless hold is legal and is discharged by a keyless release; a hold whose reason is a

package/plugins/src/base/skills/lisa-linear-access/SKILL.md

@@ -188,7 +188,7 @@ Consumers pass business-shaped arguments only; they do not embed GraphQL.
 
 **This layer does not accept a state to write. It resolves one.** A `save-issue`
 that changes the workflow state declares `lifecycle_role:<ROLE>` — the semantic
-role it is applying (`ready`, `claimed`, `blocked`, `review`, `done`) — plus `env:<KEY>` when the role is the env-indexed
+role it is applying (for example, `ready`, `claimed`, `blocked`, `review`, `done`; the configured role contract is authoritative) — plus `env:<KEY>` when the role is the env-indexed
 `done`. This layer then resolves that role against config and the team's own
 catalog, and sends the ID **it** resolved.
 
@@ -319,18 +319,67 @@ same mutation, same result shape, same error handling.
 ```graphql
 # list-comments project_id:<ID> — page via pageInfo so a long-lived PRD's
 # feedback history never silently truncates.
-query($id:String!){
+query($id:String!,$after:String){
   project(id:$id){
-    comments(first:100){
+    comments(first:100,after:$after){
       pageInfo{ hasNextPage endCursor }
-      nodes{ id body createdAt url user{ name } }
+      nodes{ id body createdAt url user{ id name } botActor{ __typename } }
     }
   }
 }
 ```
 
-The Issue form reads through `issue(id:$id){ comments{...} }` with the identical
-node shape, so a caller consumes one result format for either anchor.
+The Issue form uses `issue(id:$id)` with the same `first:100,after:$after`
+arguments and node shape. For **both** anchors, use this access-layer adapter with
+`linear_graphql` above; it accumulates every page and prints no partial history.
+A read succeeds only after `hasNextPage` is false. Errors, malformed pages, and
+missing or repeated continuation cursors fail the whole read.
+Preserve `botActor` with each comment: its non-null presence identifies automation,
+even if `user.id` is trusted. When normalizing comments, carry it forward or set
+`isBot: true`; never discard this metadata before release authorization.
+
+```bash
+linear_list_comment_pages() {  # issue|project id -> complete JSON node array
+  local anchor="$1" id="$2" after=null nodes='[]' seen='[]'
+  local query variables response connection cursor
+  case "$anchor" in issue|project) ;; *) echo "Error: invalid comment anchor" >&2; return 1 ;; esac
+  query='query($id:String!,$after:String){'"$anchor"'(id:$id){comments(first:100,after:$after){pageInfo{hasNextPage endCursor} nodes{id body createdAt url user{id name} botActor{__typename}}}}}'
+  while :; do
+    variables=$(jq -cn --arg id "$id" --argjson after "$after" '{id:$id,after:$after}') || return 1
+    response=$(linear_graphql "$query" "$variables") || return 1
+    connection=$(printf '%s' "$response" | jq -ce --arg anchor "$anchor" '
+      if (.errors != null and .errors != []) then error("GraphQL comments read failed")
+      else .data[$anchor].comments end |
+      if (.nodes | type) != "array" or (.pageInfo.hasNextPage | type) != "boolean"
+      then error("Incomplete comments page") else . end') || return 1
+    nodes=$(jq -cn --argjson prior "$nodes" --argjson page "$connection" '$prior + $page.nodes') || return 1
+    if [ "$(printf '%s' "$connection" | jq -r '.pageInfo.hasNextPage')" = false ]; then
+      printf '%s\n' "$nodes"
+      return 0
+    fi
+    cursor=$(printf '%s' "$connection" | jq -ce '.pageInfo.endCursor | select(type == "string" and length > 0)') || {
+      echo "Error: incomplete comments history; missing cursor" >&2; return 1;
+    }
+    if jq -en --argjson seen "$seen" --argjson cursor "$cursor" '$seen | index($cursor) != null' >/dev/null; then
+      echo "Error: incomplete comments history; repeated cursor" >&2
+      return 1
+    fi
+    seen=$(jq -cn --argjson seen "$seen" --argjson cursor "$cursor" '$seen + [$cursor]') || return 1
+    after="$cursor"
+  done
+}
+```
+
+Keep the returned comment objects intact when a caller evaluates a human-gate release. The
+`user.id` is the stable author identity; `user.name` is display text and never authorizes release.
+Pass structured comments and an independently supplied `trustedHumanActorIds` allowlist to the
+shared gate helpers (`ready-role-filing` — Human-gate release authorization). A body-only projection
+loses identity and cannot discharge a hold. Preserve any explicit bot metadata from the provider;
+if identity or trusted policy is unavailable, leave the hold active.
+
+Call `linear_list_comment_pages project "$PROJECT_ID"` for `project_id`, or
+`linear_list_comment_pages issue "$ISSUE_ID"` for `issue_id`. The caller receives
+one complete node array; it must not interpret a failed command as zero comments.
 
 **Substrate.** The Linear MCP exposes comments on Issues only, so the
 `project_id` form resolves solely through the tier-1 `LINEAR_API_KEY` + GraphQL
@@ -339,7 +388,7 @@ That is not a gap in Linear: the precedence contract already puts GraphQL ahead
 of the MCP, so the preferred substrate has always been able to do this. Only
 this wrapper could not, which is what made callers fabricate an Issue to hold a
 Project's comments. If tier 1 is unavailable, fail with the layer's standard
-`Error:` result naming `save-comment project_id` — do **not** silently degrade
+`Error:` result naming the requested `save-comment project_id` or `list-comments project_id` operation — do **not** silently degrade
 into creating an Issue to comment on.
 
 ## `history` — transition history (read-only)

package/plugins/src/base/skills/lisa-linear-build-intake/SKILL.md

@@ -6,6 +6,20 @@ allowed-tools: ["Skill", "Bash"]
 
 # Linear Build Intake: $ARGUMENTS
 
+## Human-gate release authorization
+
+A release requires a trusted human author, not just matching comment text. Follow
+`ready-role-filing` — **Human-gate release authorization**: preserve tracker-supplied comment
+author IDs and bot metadata, resolve `trustedHumanActorIds` only from an explicit user instruction
+or existing human-authored trusted project policy, and pass it with structured `comments` to every
+hold classifier, reconciliation, normalization and release planner. Never derive trust from the
+comment body, a display name, the actor's own assertion, or an automation posting on its own behalf.
+Missing policy, missing/unreadable author identity, raw body strings, untrusted actors and known bots
+cannot discharge a hold. Keep the item held and report the missing authorization; do not silently
+replace these inputs with an empty history or an inferred allowlist. Authorized matching releases
+continue through the existing path and never override an independently declared caller hold.
+
+
 `$ARGUMENTS` is one of:
 
 1. A Linear team key (e.g. `ENG`) — scans that team for ready Issues.
@@ -182,13 +196,21 @@ A blocker is a **claim with a timestamp, not a fact**. It is written once and go
 
 For each pre-work candidate that is **not** in `$READY`:
 
-1. **Human gate first, and it is absolute.** An Issue carrying the configured human-needed label (`linear.labels.build.human_needed`, default `human-needed`) or a `[lisa-human-gate]` marker in its description is **never** auto-selected, whatever any probe says. Skip it and move on.
+1. **Read release comments before applying the human gate.** Pass all item comments, labels,
+   body, `trustedHumanActorIds` and configured `humanNeededLabel` to `classifyReadyCandidate(...)`. For a discharged hold,
+   call `planHumanGateRelease({ labels, body, comments, trustedHumanActorIds, humanNeededLabel, readyLabel, lifecycleLabels, alreadyNotified })`
+   and retain its plan. Apply human-marker cleanup only as planned; defer any ready-role restoration
+   until steps 2–5 have cleared the remaining blockers (map the role to the vendor's state or label).
+   A still-active hold is never auto-selected, whatever any probe says. An absent or unreadable
+   comment history cannot discharge a hold. A release removes only the human hold: continue the
+   remaining blocker checks below before selecting the item.
+
 2. **Extract the stated discharge condition** from the description or the most recent blocking comment — the sentence naming what has to become true.
 3. **Probe it.** Machine-testable conditions are the ones that rot fastest and are cheapest to check: a version on trunk (`git show origin/<trunk>:<manifest>`), a published package, a run history (`gh run list`), an advisory's patched status. A condition that is a human decision is not machine-testable — leave it and move on.
 4. **Classify with the shared helper** so the ordering and the evidence requirement cannot drift per vendor:
 
 ```text
-classifyPreWorkCandidate({ laneType, labels, body, humanNeededLabel, statedBlocker, probe })
+classifyPreWorkCandidate({ laneType, labels, body, comments, trustedHumanActorIds, humanNeededLabel, statedBlocker, probe })
   → { selectable, reason, humanGated, evidence }
 ```
 
@@ -216,7 +238,7 @@ other gate's verdict — however conclusive — may promote an item a person par
    drift, and a drifted gate fails *silently*, by quietly ceasing to match. Do **not** re-implement
    the test here, and do **not** key it on `reason=`: markers in the wild carry no `reason=` key at
    all and sit anywhere in the body, so a structured parse would miss them while appearing to work
-   on every item that happens to have one. **Pass the item's `comments` alongside its labels and
+   on every item that happens to have one. **Pass the item's structured `comments` and `trustedHumanActorIds` alongside its labels and
    body.** A hold is ended by a release recorded in a comment, so a reader handed no comments cannot
    see the discharge — it goes on holding an item whose question was answered weeks ago, which is
    the defect this gate carried from the day it was written (CodySwannGT/lisa#3852). Omitting them
@@ -227,16 +249,17 @@ other gate's verdict — however conclusive — may promote an item a person par
    and re-rejected every cycle forever and seen by nothing — `lisa-repair-intake` sweeps items that
    are **not** in the ready role and excludes gated ones outright, so a ready-and-gated item falls
    outside its filter twice over. Call
-   `planHumanGateReconciliation({ labels, body, humanNeededLabel, readyLabel, alreadyNotified })`
+   `planHumanGateReconciliation({ labels, body, comments, trustedHumanActorIds, humanNeededLabel, readyLabel, alreadyNotified })`
    and apply exactly the actions it returns: remove `$READY`, add the configured human-needed
    marker, and post `formatHumanGateNote()` once. The planner is idempotent by state, so an item
    already out of the lane and already marked yields no second mutation and no second comment. This
    is the same repair the leaf-only gate already performs for a ready item that must not be
    dispatched.
 4. **On `claimable: true` for an item that still carries a hold, RELEASE it — do not just proceed.**
-   The hold left durable state behind: the item is out of the queue and flagged as needing a person,
-   and answering the question does not undo either. Call
-   `planHumanGateRelease({ labels, body, comments, humanNeededLabel, readyLabel, lifecycleLabels, alreadyNotified })`
+   This ready-lane candidate can retain a historical body marker and a human-needed label.
+   This step reconciles only candidates still in the ready lane; released holds outside it are
+   recovered by `lisa-repair-intake` step 2b (or Phase 2.5 when included in this cycle). Call
+   `planHumanGateRelease({ labels, body, comments, trustedHumanActorIds, humanNeededLabel, readyLabel, lifecycleLabels, alreadyNotified })`
    and apply exactly the actions it returns: remove the configured human-needed marker, add the
    configured ready role back, and post `formatHumanGateReleaseNote()` once. It is the exact inverse
    of step 3's planner and it refuses in both directions — an item still held plans nothing, and an
@@ -268,10 +291,12 @@ A Linear team can oversee multiple repos (`frontend` / `backend` / `infrastructu
 1. **Resolve the current repo** per `config-resolution` "Repo scoping" (`.repo` → `.github.repo` → `git remote get-url origin` basename). If unresolvable, stop and report.
 2. **Cheap path first.** Prefer candidates already carrying the `repo:<current>` label. Keep the Phase 2 scan broad so unlabeled Issues are still seen, determined, and stamped.
 3. **Per candidate, apply the repo-scope decision (`repo-scope-split`):**
-   - Carries `repo:<other>` → **skip** (leave it `ready` for that repo's own intake); next candidate.
-   - **Unlabeled** → determine the target repo(s) from the Issue + code surfaces, then **stamp** `repo:<name>` via `lisa-linear-access operation: save-issue` (resolve/create the label via `list_issue_labels`/`create_issue_label`) so later cycles filter cheaply; re-apply with the now-known repo.
-   - **Multi-repo leaf → split, never claim.** Run the `repo-scope-split` work-time procedure into single-repo siblings, each created **build-ready** (`build_ready: true`) and stamped with its own `repo:<name>`; the current repo's sibling becomes a normal candidate.
-   - **Single-repo leaf for the current repo** → fall through to 3a (leaf-only gate) and 3b (claim).
+   - **Count distinct repository markers first**: collect all `repo:<name>` labels and, on JIRA, recognized repository components; deduplicate by repository. A container may carry multiple `repo:<name>` labels. Do not split or claim it here; send it to the leaf-only gate.
+   - **Multi-repo leaf → split, never claim.** More than one repository takes the work-time split before any wrong-repository skip. Each sibling is **build-ready** (`build_ready: true`) and stamped with its own `repo:<name>`; the current repo's sibling becomes a normal candidate.
+   - **Exactly one other repository** (`repo:<other>`) → **skip**, leaving the item ready for that repo's intake.
+   - **No repository marker** → determine target repo(s) from the ticket and code, stamp `repo:<name>` through the vendor access layer, and re-apply from the count.
+   - **Single-repo leaf for the current repo** → fall through to 3a and 3b.
+
 4. Continue until a claimable current-repo leaf is found (claim it; one per cycle) or the candidate set is exhausted — exit cleanly on the denominator-stated summary, naming the current repo alongside the swept lanes.
 
 #### 3a. Leaf-only claim gate (skip / safe-block containers)

package/plugins/src/base/skills/lisa-linear-prd-intake/SKILL.md

@@ -130,6 +130,16 @@ Select the first ready project returned by Phase 2 and process only that project
 
 #### 3a. Claim
 
+**Preflight project comments before any claim or destination-ticket creation.** Through
+`lisa-linear-access`, resolve an identity-matched tier-1 GraphQL credential and the
+`save-comment project_id` adapter, then read the candidate's complete project comment
+history with `list-comments project_id:<id>`. The issue-only MCP is insufficient for
+this flow. If the required credential, adapter or read is unavailable, return the
+layer's `Error:` naming the project-comment operation and leave the project ready;
+create no tickets and no substitute comment-holder Issue. This is a read-only
+capability preflight, not a test comment. It establishes the available transport;
+later permission changes or write failures still follow the normal error handler.
+
 Transition labels via `lisa-linear-access operation: save-project({id, labels})`: pass the full new label set with `$READY` removed and `$IN_REVIEW` added. This is the idempotency lock — a re-entrant cycle running concurrently won't see this project because its query filters on `label: "$READY"`.
 
 If the update fails (permission error, race condition), log it and skip this project. Do not proceed to validation on a project you didn't successfully claim.
@@ -152,7 +162,7 @@ This call also indirectly invokes `lisa-tracker-source-artifacts` (artifact extr
 
 1. Re-invoke `lisa-linear-to-tracker` with `dry_run: false` to actually write the tickets. This re-runs Phases 1-5 and runs the preservation gate (Phase 5.5).
 2. Capture the created ticket keys from the skill's output.
-3. Post a comment on the project via `lisa-linear-access operation: save-comment project_id:<id>` listing the created tickets (epic, stories, sub-tasks) with their JIRA URLs. Lead with: `"Ticketed by Claude. Created N JIRA issues — see below. Add the $SHIPPED label to the Linear project after the work is delivered."`
+3. Post a comment on the project via `lisa-linear-access operation: save-comment project_id:<id>` listing the created tickets (epic, stories, sub-tasks) with the URLs returned by the configured destination tracker. Lead with: `"Ticketed by Claude. Created N tickets — see below. Add the $SHIPPED label to the Linear project after the work is delivered."`
 4. Transition labels: remove `$IN_REVIEW`, add `$TICKETED` via `save_project`.
 5. **Run Phase 3e (coverage audit)** before considering this PRD done.
 
@@ -396,6 +406,6 @@ Before this skill can run against a Linear workspace or team, the team must adop
 1. Apply the `ready` label (default: `prd-ready`) to projects that are ready for ticketing (replaces the Notion `Status = Ready` flip and the Confluence `prd-ready` page label).
 2. Reserve `in_review`, `blocked`, `ticketed` (defaults: `prd-in-review`, `prd-blocked`, `prd-ticketed`) for this skill — humans should not set them manually except to recover from an error.
 3. (Optional but recommended) Add the `draft` and `shipped` labels (defaults: `prd-draft`, `prd-shipped`) for in-progress PRDs and delivered work respectively, so the full lifecycle is visible at a glance.
-4. The labels must exist as **project labels** in Linear (`list_project_labels` should return them). Issue-level labels with the same names won't work; Linear keeps the two label kinds separate.
+4. The labels must exist as **project labels** in Linear (`list-project-labels` should return them). Issue-level labels with the same names won't work; Linear keeps the two label kinds separate.
 
 If the workspace hasn't adopted these labels, the first run exits with a label-convention error (not the idle empty-set message) — this distinguishes a setup issue from a genuinely empty queue so operators know to apply the convention rather than assuming there is no work. See Phase 2 for how the skill detects this case.

package/plugins/src/base/skills/lisa-linear-validate-issue/SKILL.md

@@ -91,24 +91,7 @@ prd_source: "https://notion.so/..."    # set when the Issue was generated from a
 
 If the caller passes only an identifier, fetch the item via `lisa-linear-access operation: get-issue` (Issue) or `lisa-linear-access operation: get-project` (Project), derive the same fields from the fetched data — including `runtime_behavior_change` (derived from the `## Target Backend Environment` declaration per `derived-branch-plan`, and authoritative over any caller assertion), `build_ready` (the Issue's state is the configured `ready` state) and `child_refs` (sub-issues, project-member issues, plus `blocked_by` parentage, resolved as in `lisa-linear-read-issue`) so S15 can classify the item — then run gates.
 
-## Standalone use, against an item that already exists
-
-This is a supported entry point, not only an internal step of a caller flow.
-Point it at a live issue and it fetches and validates the stored state:
-
-```text
-Skill(lisa-linear-validate-issue) with ENG-1234
-```
-
-Use it whenever the issue reached the tracker by some path other than
-`lisa-linear-write-issue` — a team's own script, a workflow step, a cron, anything holding the
-credentials but no skill runtime. Those paths get neither the pre-write nor the
-post-write gate, and their own read-back substitutes for neither: a read-back
-proves the tracker stored what was sent, never that what was sent was any good.
-
-The gate definitions live here on purpose, so every caller picks up a change
-automatically. Running this skill by hand is the same gate the write path runs,
-not an approximation of it.
+For existing items written through another path, use the [standalone entry point](#standalone-entry-point--validating-an-item-written-by-another-path). A transport read-back does not prove the stored specification passes the gates.
 
 ## Gates
 

package/plugins/src/base/skills/lisa-linear-verify/SKILL.md

@@ -63,20 +63,3 @@ normalize-then-compare. Byte-exact comparison of such text is forbidden.
 - Never write to Linear. Read-only.
 - Never short-circuit the validator. Always run the full gate set.
 - If `get_issue` / `get_project` returns an access error, surface it and exit — don't pretend the item is fine.
-
-## Comparison is semantic, never byte-exact
-
-Re-run the validator against the live issue. Do NOT compare the stored body
-against what was sent byte for byte.
-
-The reason is measured rather than theoretical. Trackers normalize markdown on
-write: `-` bullets become `*`, a bare URL is wrapped as `[url](<url>)`, bold
-emphasis is re-segmented around inline code spans. All lossless, all
-rendering-identical, and all of it makes a byte comparator report failure on a
-write that was completely fine. A comparator that cannot tell vendor
-normalization from corruption fails on healthy writes and trains its reader to
-ignore it, which costs more than the check was ever worth.
-
-Compare meaning: run `lisa-linear-validate-issue` against the stored item and let the gates
-decide. Where a single field must be compared directly, normalize both sides
-first.