@codyswann/lisa 4.45.2 → 4.45.4

package/plugins/lisa/hooks/block-direct-issue-create.sh

@@ -346,7 +346,7 @@ session found 13 of 13 issues filed this way, none of them through Lisa's
 filing path. The rule saying not to do this already existed — it did not bind,
 so it is now enforced here.
 
-FILE IT THE SANCTIONED WAY — one of these two, always explicit:
+FILE IT THE SANCTIONED WAY — one of these three, always explicit:
 
 1. The item is complete enough to build. Use the filing flow, not the CLI:
 
@@ -368,11 +368,26 @@ FILE IT THE SANCTIONED WAY — one of these two, always explicit:
      Held for a human product call: <reason>.
      <!-- [lisa-human-gate] reason=<short-slug> -->
 
+3. The item is a CONTAINER — an Epic, or any item whose state rolls up from
+   its children. It is neither of the two above and must not pretend to be:
+   \`leaf-only-lifecycle\` FORBIDS the build-ready role on a container, and a
+   human gate it does not have is a durable false hold. It declares the third
+   thing instead — the canonical container line, in place of a Target Backend
+   Environment:
+
+     ## Target Backend Environment
+
+     None — container: state rolls up from children
+
+   That declaration is refused alongside a by-design leaf type
+   (\`type:Bug\` / \`type:Task\` / \`type:Sub-task\` / \`type:Improvement\`):
+   an item cannot be a container and a leaf at once.
+
 Filed, not ready, and no \`human_gate\` is the incomplete-handoff case, and
 \`build_ready: false\` with no reason is the same omission with a value
 attached. See the \`ready-role-filing\` rule for the full contract.
 
-If you must run the CLI directly, the command has to carry one of the two
+If you must run the CLI directly, the command has to carry one of the three
 declarations itself:
 
 $(declaration_hint)
@@ -380,6 +395,9 @@ $(declaration_hint)
    Or a \`[lisa-human-gate]\` marker in the body it submits, when a human
    product call really is pending.
 
+   Or, for a container, \`None — container: state rolls up from children\` in
+   that body, and no by-design leaf type on the command.
+
 WHERE THE DECLARATION IS READ FROM: argv, the request payload — inline, in a
 \`--data-binary @file\`, or piped in over stdin — and the contents of a script
 this command runs. Moving the create into a file no longer moves it out of
@@ -528,6 +546,78 @@ UNPARSEABLE_CREATION = re.compile(
 
 BODY_FILE_FLAGS = {"--body-file", "-F", "--input", "--data-binary"}
 LABEL_FLAGS = {"--label", "--labels", "--add-label", "--status", "--state"}
+
+# ---------------------------------------------------------------------------
+# THE THIRD DECLARATION: THIS ITEM IS A CONTAINER
+#
+# The two declarations above are both illegitimate for a container, and the
+# rules say so in as many words. `leaf-only-lifecycle` FORBIDS the build-ready
+# role on an Epic — its state rolls up from its children, and a hand-applied
+# role on a parent is the exact input that rule's claim-time arm exists to
+# reject. `lisa-github-write-issue` says a container needs no human gate.
+# So a guard with only those two arms could be satisfied for a container only
+# by writing something untrue, and BOTH untruths corrupt data another control
+# reads: a stamped role puts a container in the build lane, and a fabricated
+# hold marker is a state change with no inverse that reads to every later
+# observer as a real product hold. A guard that can only be obeyed by lying is
+# worse than one that simply refuses, because the lie is indistinguishable from
+# a correct filing afterwards.
+#
+# WHY NOT `--label type:Epic`
+#
+# Because a declared type is a CLAIM, and one that costs a leaf nothing: the
+# item it produces still looks buildable, so the exemption would be a free
+# bypass of the readiness requirement for anything willing to mistype itself.
+# That is the shape of this repository's measured case where an allowlist added
+# to harden a guard became the way around it.
+#
+# WHAT IS READ INSTEAD
+#
+# The canonical container declaration the `derived-branch-plan` rule already
+# defines and `lisa-github-write-issue` already stamps, in place of a Target
+# Backend Environment:
+#
+#   None — container: state rolls up from children
+#
+# It is a marker with no other meaning, so it is matched wherever the payload
+# reaches — exactly the asymmetry `HUMAN_GATE_MARKER` gets, and for the same
+# reason. Reading the declaration the writer already emits is also what keeps
+# the guard and the skill from drifting apart a second time: there is one
+# string, defined in one rule, and both read it.
+#
+# It is self-limiting in a way a type label is not. Writing it costs the item
+# its Target Backend Environment and its Branch Plan — the two fields a leaf
+# needs to be built at all — so the declaration is only useful to something
+# that actually is a container. A leaf that writes it does not smuggle a
+# buildable item past the gate; it files an unbuildable one, in the lane
+# `leaf-only-lifecycle` and `lisa-repair-intake` already sweep.
+#
+# The WHOLE canonical value, `None` included, not just the distinctive tail.
+# Dropping the leading half would match ordinary prose about how a container's
+# state behaves — including a bug report ABOUT this guard — and a declaration
+# that ordinary prose satisfies is not a declaration. Whitespace-tolerant
+# because a body file may wrap the line; dash-tolerant because an author
+# retyping it by hand will not always reach for the em dash.
+CONTAINER_DECLARATION = re.compile(
+    r"None\s*[\u2014\u2013-]\s*container:\s*state\s+rolls\s+up\s+from\s+children",
+    re.IGNORECASE,
+)
+# The checkable half of "is it really a container". Provenance is unobservable
+# here, as the header notes, but a CONTRADICTION is not: an item cannot be a
+# container and a by-design leaf at once. Quoted from `leaf-only-lifecycle` —
+# "the by-design leaf types (Bug, Task, Sub-task, Improvement)". Story and
+# Spike are deliberately ABSENT: the same rule's childless-parent exception
+# makes them leaf-or-container depending on child work, and a decomposition
+# legitimately files a parent Story before the children that make it one.
+BY_DESIGN_LEAF_TYPES = {"bug", "task", "sub-task", "subtask", "improvement"}
+# Where a declared type can land on the created item. The label spellings are
+# GitHub's `type:<value>` convention; `--type` and its aliases are the JIRA and
+# Linear clients' spelling. Read with the same positional discipline as the
+# build-ready role — a type named in a title or body is not a type applied.
+TYPE_FLAGS = {
+    "--label", "--labels", "--add-label",
+    "--type", "--issue-type", "--issuetype",
+}
 POST_METHOD_FLAGS = {"-X", "--request", "--method"}
 POST_PAYLOAD_FLAGS = {
     "-d", "--data", "--data-raw", "--data-binary",
@@ -1433,6 +1523,65 @@ def declares_readiness(raw_args, roles, extra="", state_role_ok=False):
     return False
 
 
+def declares_leaf_type(args):
+    """Whether the filing declares a by-design leaf type as a flag value.
+
+    Read only from positions that reach the created item, and never from a
+    title or body — the same discipline the build-ready role is held to, for
+    the same reason: a type named in prose is not a type applied.
+
+    Args:
+        args: The creating command's arguments, already scoped to the flags.
+
+    Returns:
+        True when a Bug / Task / Sub-task / Improvement is declared.
+    """
+    for raw in flag_values(args, TYPE_FLAGS):
+        for part in raw.split(","):
+            value = part.strip().strip("'\"").lower()
+            if value.startswith("type:"):
+                value = value[len("type:") :]
+            if value in BY_DESIGN_LEAF_TYPES:
+                return True
+    return False
+
+
+def declares_container(raw_args, texts=()):
+    """Whether the create declares the item a container.
+
+    A container is neither build-ready nor human-gated by construction, so this
+    is the third accepted declaration rather than a third way to satisfy the
+    first two. See the CONTAINER_DECLARATION block for why it reads the
+    canonical declaration and not a `type:Epic` claim.
+
+    Args:
+        raw_args: The creating command's arguments.
+        texts: Payload or script text this command submits or runs.
+
+    Returns:
+        True when the container declaration is present and uncontradicted.
+    """
+    args = before_end_of_options(raw_args)
+    # The contradiction check comes FIRST, so a filing that declares itself
+    # both a container and a by-design leaf is refused rather than allowed on
+    # the strength of the half that suits it.
+    if declares_leaf_type(args):
+        return False
+    if CONTAINER_DECLARATION.search(" ".join(args)):
+        return True
+    for path in body_file_paths(args):
+        try:
+            with open(path, encoding="utf-8", errors="replace") as handle:
+                if CONTAINER_DECLARATION.search(handle.read()):
+                    return True
+        except OSError:
+            continue
+    for text in texts:
+        if text and CONTAINER_DECLARATION.search(text):
+            return True
+    return False
+
+
 def flag_values(args, names):
     """Every value assigned to one of the named flags.
 
@@ -1880,6 +2029,15 @@ def scan(text, depth, from_file=False):
             state_role_ok = tracker in STATE_ROLE_TRACKERS and target is None
             if declares_readiness(args, roles, submitted, state_role_ok):
                 continue
+            # The container arm. Checked after readiness, so it adds a
+            # compliant filing where none existed and takes none away where one
+            # already did. `text` joins the scan only for a creation found
+            # INSIDE a file, where the declaration lives in the same file as
+            # the create it answers for.
+            if declares_container(
+                args, (submitted, text) if from_file else (submitted,)
+            ):
+                continue
             # `LIFECYCLE_ROLE=ready curl …` puts the declaration BEFORE the
             # client, which is where an inline assignment has to go, so the
             # role is read from the whole segment rather than from the client's

package/plugins/lisa/hooks/parity-safety-net.sh

@@ -405,7 +405,12 @@ readonly FOLLOW_MAX_TOTAL_BYTES=524288
 readonly FOLLOW_MAX_FILES=8
 readonly FOLLOW_MAX_DEPTH=3
 
-follow_bases=""
+# Effective working directory established by a literal `cd` in the command, and
+# the flag that says a `cd` happened whose target could NOT be read literally.
+# Empty + 0 means no `cd` has been seen and the session's own cwd governs.
+follow_cwd=""
+follow_cwd_unknown=0
+follow_normalized=""
 follow_paths=$'\n'
 follow_names=""
 follow_count=0
@@ -420,6 +425,10 @@ see is the thing it exists to stop.
 
 - Name the script by a LITERAL path that exists and is readable — \`bash
   ./scripts/x.sh\` rather than \`bash \"\$SCRIPT\"\` — and the guard can classify it.
+- A leading \`cd\` the guard cannot read literally — \`cd \"\$DIR\" && bash x.sh\` —
+  leaves it unable to say WHICH copy of \`x.sh\` would run, and a same-named file
+  in another directory is a different file. Spell the directory out, or give the
+  script its full path.
 - A file past the inspection cap is refused rather than half-scanned, because a
   truncated scan reports a confident ALLOW about text it never read. Split it,
   or run it manually outside the agent.
@@ -550,15 +559,77 @@ follow_split() {
   '
 }
 
-# Record a literal `cd` target as an additional base for resolving a relative
-# executed path, so `cd sub && ./run.sh` resolves the way the shell will.
-follow_add_base() {
-  local d="$1"
+# Collapse `.`, `..` and empty segments so the path a verdict NAMES is the one a
+# reader recognises. Purely lexical, and it cannot move a verdict: every
+# component has already been proven to exist by the `-d` test that follows, and
+# the `-f`/`-r` tests are applied to the collapsed path either way.
+follow_normalize() {
+  local p="$1" out="" seg
+  local IFS=/
+  set -f
+  # The word split is the point: `$p` is being addressed one path segment at a
+  # time, and the glob guard above keeps a literal `*` from expanding.
+  # shellcheck disable=SC2086
+  for seg in $p; do
+    case "$seg" in
+      "" | .) ;;
+      ..) out="${out%/*}" ;;
+      *) out="$out/$seg" ;;
+    esac
+  done
+  set +f
+  follow_normalized="${out:-/}"
+}
+
+# Track the working directory a literal `cd` establishes, so a relative executed
+# path resolves against the directory the command will actually have.
+#
+# This REPLACES the session's own cwd rather than adding to it. It used to be an
+# extra base appended AFTER `$PWD`, so `$PWD` always won and
+# `cd <other-checkout> && bash scripts/x.sh` was classified against the session
+# checkout's same-named copy. That is not a resolution failure — it succeeds,
+# against a real and readable file, and the verdict is reported with full
+# confidence about text nobody was about to run (CodySwannGT/lisa#3933). Two
+# checkouts of one repository differ by exactly the work in flight, which is the
+# entire point of the second one, so the copy that would RUN is the only copy
+# whose verdict means anything. Wrong in both directions, and the expensive
+# direction is silent: the dangerous line sits in the copy that would run and
+# the guard reads the pristine one.
+#
+# A target this hook cannot turn into a literal directory marks the cwd UNKNOWN
+# instead of leaving a stale one in place. `follow_locate` then refuses a
+# relative token rather than guessing, which is the honest-refusal doctrine the
+# rest of this block already follows: a confident scan of the wrong file is
+# strictly worse than saying it could not tell.
+follow_cd() {
+  local d="$1" home="${HOME:-}"
+  # The tilde arms are LITERAL text — the unexpanded spelling the hook was
+  # handed — for the reason follow_locate's arms document.
+  # shellcheck disable=SC2088
+  case "$d" in
+    '~') d="$home" ;;
+    '~/'*) d="$home/${d#'~/'}" ;;
+  esac
+  case "$d" in
+    "" | -* | *'$'* | *'*'* | *'?'* | *'`'* | *'{'* | *'['*)
+      follow_cwd=""
+      follow_cwd_unknown=1
+      return 0
+      ;;
+  esac
   case "$d" in
-    "" | -* | *'$'* | *'*'* | *'?'*) return 0 ;;
+    /*) ;;
+    *) d="${follow_cwd:-$PWD}/$d" ;;
   esac
-  [ -d "$d" ] || return 0
-  follow_bases="$follow_bases"$'\n'"$d"
+  follow_normalize "$d"
+  d="$follow_normalized"
+  if [ -d "$d" ]; then
+    follow_cwd="$d"
+    follow_cwd_unknown=0
+  else
+    follow_cwd=""
+    follow_cwd_unknown=1
+  fi
 }
 
 # Resolve one executed-file operand to a readable regular file, or fail.
@@ -571,6 +642,7 @@ follow_add_base() {
 # this hook cannot evaluate.
 follow_locate() {
   local token="$1" home="${HOME:-}" tmp="${TMPDIR:-/tmp}" proj="${CLAUDE_PROJECT_DIR:-$PWD}" base
+  local cwd="${follow_cwd:-$PWD}"
   follow_located=""
   tmp="${tmp%/}"
   [ -n "$tmp" ] || tmp="/tmp"
@@ -584,8 +656,14 @@ follow_locate() {
     '${TMPDIR}' | '${TMPDIR}'/*) token="$tmp${token#'${TMPDIR}'}" ;;
     '$HOME' | '$HOME'/*) token="$home${token#'$HOME'}" ;;
     '${HOME}' | '${HOME}'/*) token="$home${token#'${HOME}'}" ;;
-    '$PWD' | '$PWD'/*) token="$PWD${token#'$PWD'}" ;;
-    '${PWD}' | '${PWD}'/*) token="$PWD${token#'${PWD}'}" ;;
+    '$PWD' | '$PWD'/*)
+      [ "$follow_cwd_unknown" -eq 0 ] || return 1
+      token="$cwd${token#'$PWD'}"
+      ;;
+    '${PWD}' | '${PWD}'/*)
+      [ "$follow_cwd_unknown" -eq 0 ] || return 1
+      token="$cwd${token#'${PWD}'}"
+      ;;
     '$CLAUDE_PROJECT_DIR' | '$CLAUDE_PROJECT_DIR'/*) token="$proj${token#'$CLAUDE_PROJECT_DIR'}" ;;
     '${CLAUDE_PROJECT_DIR}' | '${CLAUDE_PROJECT_DIR}'/*) token="$proj${token#'${CLAUDE_PROJECT_DIR}'}" ;;
   esac
@@ -597,13 +675,26 @@ follow_locate() {
       return 0
       ;;
   esac
+  # Past this point the token is RELATIVE, so which directory it is resolved
+  # from decides which file the verdict is about.
+  if [ "$follow_cwd_unknown" -eq 1 ]; then
+    return 1
+  fi
+  if [ -n "$follow_cwd" ]; then
+    # A literal `cd` names the directory the shell will be in. It is the only
+    # candidate — falling back to `$PWD` here is what made the guard report
+    # confidently about a same-named file in another checkout.
+    [ -f "$follow_cwd/$token" ] && [ -r "$follow_cwd/$token" ] || return 1
+    follow_located="$follow_cwd/$token"
+    return 0
+  fi
   while IFS= read -r base; do
     [ -n "$base" ] || continue
     if [ -f "$base/$token" ] && [ -r "$base/$token" ]; then
       follow_located="$base/$token"
       return 0
     fi
-  done <<<"$PWD"$'\n'"$proj$follow_bases"
+  done <<<"$PWD"$'\n'"$proj"
   return 1
 }
 
@@ -636,7 +727,15 @@ follow_take() {
   follow_names="$follow_names $path"
   followed_text="$followed_text"$'\n'"$content"
   if [ "$depth" -lt "$FOLLOW_MAX_DEPTH" ]; then
+    # A `cd` inside the followed script governs what THAT script goes on to
+    # execute, and nothing after it in the command that ran the script — the
+    # child inherits this directory but cannot change the parent's. Saving and
+    # restoring is what keeps a nested `cd` from silently re-pointing the rest
+    # of the outer walk.
+    local saved_cwd="$follow_cwd" saved_unknown="$follow_cwd_unknown"
     follow_scan "$content" "$((depth + 1))"
+    follow_cwd="$saved_cwd"
+    follow_cwd_unknown="$saved_unknown"
   fi
 }
 
@@ -861,7 +960,11 @@ follow_scan() {
         cd)
           if [ $((i + 1)) -lt "$n" ]; then
             follow_unwrap "${toks[i + 1]}"
-            follow_add_base "$follow_unwrapped"
+            follow_cd "$follow_unwrapped"
+          else
+            # A bare `cd` is `cd "$HOME"`, and leaving the previous directory in
+            # place would resolve later tokens from a directory the shell left.
+            follow_cd "${HOME:-}"
           fi
           cmd_pos=0
           i=$((i + 1))

package/plugins/lisa/rules/eager/ready-role-filing.md

@@ -6,10 +6,11 @@ Filing a work item without the ready role is an **incomplete handoff** — build
 
 **An omitted `build_ready` is NOT build-ready on JIRA, GitHub, and Linear alike.** Ready is an explicit claim, never an accident of which tracker a project configured. This is a **breaking change** for GitHub and Linear callers, which previously treated omission as ready; `lisa-github-validate-issue` F4's compensating omitted → `true` normalization is removed with it.
 
-## Every filing declares one of two things
+## Every filing declares one of three things
 
 - **`build_ready: true`** — complete enough to build; enters the ready lane for auto-pickup. Required for any complete defect found during other work, so it is claimable next cycle with no human flipping status.
 - **`human_gate: "<why a human must judge this first>"`** — deliberately held outside the queue. The writer stamps a visible line plus `<!-- [lisa-human-gate] reason=<short-slug> -->` so the hold is auditable.
+- **The container declaration** — `None — container: state rolls up from children`, the canonical line `derived-branch-plan` defines and every writer stamps in place of a Target Backend Environment. A container is never build-ready (`leaf-only-lifecycle`) and needs no gate, so it declares the third thing instead. Deliberately **not** keyed on `--label type:Epic`: a declared type is a claim that costs a leaf nothing and leaves it looking buildable, where the container line costs the item the environment and Branch Plan a leaf needs to be built. It is refused alongside a by-design leaf type (`type:Bug` / `type:Task` / `type:Sub-task` / `type:Improvement`) — an item cannot be a container and a leaf at once.
 
 Filed, not ready, and no `human_gate` is the **incomplete handoff** case — writers reject it and name both ways to resolve it. `build_ready: false` with no reason is the same omission with a value attached. `build_ready` stays subordinate to `leaf-only-lifecycle`: a container is never build-ready and needs no gate.
 
@@ -21,7 +22,7 @@ Filed, not ready, and no `human_gate` is the **incomplete handoff** case — wri
 
 ## This rule is enforced
 
-The PreToolUse guard `block-direct-issue-create.sh` refuses a direct tracker-creation command (`gh issue create`, `gh api` POST to an issues endpoint, `linear`/`jira`/`acli` creates, equivalent `curl` posts) that carries **no readiness declaration** — no configured build-ready role and no `[lisa-human-gate]` marker. It shipped because this rule as prose did not bind: 13 of 13 issues filed during the session that merged it bypassed it, several by the agent that wrote it, while the one obligation backed by a git hook was honored 50 of 50 times. It stands down where no tracker is configured, and for an operator who set `LISA_ALLOW_DIRECT_ISSUE_CREATE` **in the ambient environment** — never from an inline assignment on the refused command.
+The PreToolUse guard `block-direct-issue-create.sh` refuses a direct tracker-creation command (`gh issue create`, `gh api` POST to an issues endpoint, `linear`/`jira`/`acli` creates, equivalent `curl` posts) that carries **none of the three declarations** — no configured build-ready role, no `[lisa-human-gate]` marker, and no container declaration. It shipped because this rule as prose did not bind: 13 of 13 issues filed during the session that merged it bypassed it, several by the agent that wrote it, while the one obligation backed by a git hook was honored 50 of 50 times. It stands down where no tracker is configured, and for an operator who set `LISA_ALLOW_DIRECT_ISSUE_CREATE` **in the ambient environment** — never from an inline assignment on the refused command.
 
 **The role is the TARGET repository's, not the caller's.** A creation addressed at another repository — `--repo` / `-R`, or a `repos/<org>/<repo>/issues` endpoint — is judged against that repository's build-ready role: `hardening.upstreamReadyRole` (default `status:ready`) when the target is `hardening.upstreamRepo`, and the stock `status:ready` otherwise. A creation naming no repository, or naming the project's own `github.org`/`github.repo`, is judged against the project's own role exactly as before. A declaration is required either way; only its vocabulary changes.
 

package/plugins/lisa/rules/reference/ready-role-filing.md

@@ -22,7 +22,7 @@ So the same vendor-neutral `lisa-tracker-write` call produced a different lifecy
 
 The safe direction is the implicit one. A ticket that reaches a build queue by accident is worse than one that waits, because the queue is what agents act on autonomously.
 
-## Every filing declares one of two things
+## Every filing declares one of two things — three, counting containers
 
 A writer accepts a filing only when it carries **one** of:
 
@@ -38,6 +38,24 @@ Filed, not ready, and no `human_gate` is the **incomplete handoff** case: the wr
 
 `build_ready` remains strictly subordinate to `leaf-only-lifecycle`: a container is never build-ready regardless of what a caller passes, and a container needs no `human_gate` because its state rolls up from its children rather than being claimed.
 
+### The container declaration, on the direct-CLI path
+
+That exemption used to live only in writer prose, which meant the guard below — the thing that actually runs — had no arm for it. A container could satisfy the guard only by declaring a build-ready role `leaf-only-lifecycle` forbids on it, or a human gate it does not have. **A guard that can only be obeyed by writing something untrue is worse than one that simply refuses**, because the lie is indistinguishable from a correct filing afterwards, and both available lies corrupt data another control reads: a stamped role puts a container in the lane build-intake claims from, and a fabricated hold marker is a state change with no inverse.
+
+So a container filed directly through a tracker CLI declares a **third** thing — the canonical container line the `derived-branch-plan` rule already defines and every writer already stamps in place of a Target Backend Environment:
+
+```text
+## Target Backend Environment
+
+None — container: state rolls up from children
+```
+
+**Not `--label type:Epic`.** A declared type is a claim, and one that costs a leaf nothing: the item it produces still looks buildable, so keying the exemption on it would hand any item willing to mistype itself a free bypass of the readiness requirement — the shape of the measured case where an allowlist added to harden a guard became the way around it. The container line is self-limiting instead: writing it costs the item the Target Backend Environment and Branch Plan a leaf needs to be built at all, so it is only useful to something that actually is a container.
+
+The declaration is refused alongside a **by-design leaf type** declared as a flag value (`type:Bug` / `type:Task` / `type:Sub-task` / `type:Improvement`, or the `--type` spelling on a state-based tracker) — an item cannot be a container and a leaf at once. Story and Spike are deliberately not in that set: `leaf-only-lifecycle`'s childless-parent exception makes them leaf-or-container depending on child work, and a decomposition legitimately files a parent Story before the children that make it one.
+
+One string, defined in one rule, read by the guard and stamped by the writers — which is what keeps this exemption from drifting back into prose only.
+
 ## Complete defects found during other work are filed build-ready
 
 Any defect discovered while doing something else, which is complete enough to build, is filed through `lisa-track` / `lisa-tracker-write` with **explicit `build_ready: true`**. It must be claimable by build-intake on the next cycle with no human flipping status. "Complete enough to build" is the `work-item-definition-of-ready` bar — reproduction, observed-versus-expected, and Gherkin acceptance criteria — not a placeholder to be fleshed out later.
@@ -60,7 +78,7 @@ The rule shipped as prose first, and prose did not hold. A conformance audit of
 
 The control is the PreToolUse Bash guard `block-direct-issue-create.sh`, shipped to every agent variant (Claude, Codex, Cursor, Copilot, agy, OpenCode) and to the host `scripts/lisa-hooks/` fallback. It refuses a direct tracker-creation command — `gh issue create`, `gh api` POST to an issues endpoint, `gh api graphql createIssue`, `linear issue create`, `jira issue create`, `acli … create`, and equivalent `curl` posts — when the command carries **no readiness declaration**.
 
-What it checks is the artifact, not the caller. A Bash-level hook cannot observe call provenance: any provenance signal is settable by the very agent being governed, so a guard built on one would be theatre. So the guard asks whether the command about to run produces a correctly declared item — the configured build-ready role is applied, or a `[lisa-human-gate]` marker is present (inline, or in the `--body-file` the create is about to submit). That is precisely the machine-checkable content of this rule, and it lets the three writers through by construction, because they always stamp one.
+What it checks is the artifact, not the caller. A Bash-level hook cannot observe call provenance: any provenance signal is settable by the very agent being governed, so a guard built on one would be theatre. So the guard asks whether the command about to run produces a correctly declared item — the configured build-ready role is applied, or a `[lisa-human-gate]` marker is present, or the container declaration is (each read inline, or in the `--body-file` the create is about to submit). That is precisely the machine-checkable content of this rule, and it lets the three writers through by construction, because they always stamp one.
 
 Two ways out, both deliberate:
 

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

@@ -289,6 +289,22 @@ proceed — the marker still holds. Never file the label *instead of* the marker
 
 If a leaf arrives with `build_ready` omitted or `false` **and** no `human_gate`, do not create it: report the incomplete handoff and name both ways to resolve it (`build_ready: true`, or a `human_gate` reason). Containers are exempt — their state rolls up from children, so they need neither.
 
+**The container exemption is executable, not prose.** It used to live only in the sentence above,
+which meant the `block-direct-issue-create.sh` guard — the control that actually runs on the CLI
+path — had no arm for it, and a container could satisfy that guard only by declaring a build-ready
+role `leaf-only-lifecycle` forbids on it or a human gate it does not have. A container now declares
+a **third** thing, and this skill already writes it: the canonical Target Backend Environment value
+`None — container: state rolls up from children`, defined once by `derived-branch-plan`. The guard
+reads that one string, so there is nothing for the skill and the guard to disagree about.
+
+Two consequences for a container written from here. Render the Target Backend Environment section
+with that exact value — an absent section reads as *underivable*, not exempt, and leaves the guard
+nothing to read. And do not put a **by-design leaf type** (`type:Bug` / `type:Task` /
+`type:Sub-task` / `type:Improvement`) on a filing carrying that declaration: an item cannot be a
+container and a leaf at once, and the guard refuses the combination. `type:Story` and `type:Spike`
+are not in that set, because `leaf-only-lifecycle`'s childless-parent exception makes them
+leaf-or-container depending on child work.
+
 ## Phase 5.5 — Validate (Pre-write Gate)
 
 Before any write, invoke `lisa-github-validate-issue` with the full proposed spec assembled from Phases 2 / 3 / 4 / 5. Pass it as a YAML block per the `lisa-github-validate-issue` schema, including `runtime_behavior_change`, `authenticated_surface`, and `artifacts_attached` flags so the right gates run.

package/plugins/lisa-agy/hooks/block-blind-automerge.agy.sh

@@ -7,9 +7,10 @@
 # adapter only translates protocols; all classification — which commands arm
 # auto-merge, and what GitHub says about the PR's review gate — stays in the
 # canonical block-blind-automerge.sh beside it (same delegation pattern as
-# block-shell-json-parsing.agy.sh). Fail-open on missing runtimes, matching the
-# canonical guard's own degrade-loudly path: a guard that cannot read the PR
-# cannot tell an honest arming from a blind one.
+# block-shell-json-parsing.agy.sh) — including which commands re-target a pull
+# request onto a base ref no ruleset covers. Fail-open on missing runtimes,
+# matching the canonical guard's own degrade-loudly path: a guard that cannot
+# read the PR cannot tell an honest arming from a blind one.
 set -uo pipefail
 
 allow() {
@@ -50,5 +51,5 @@ else
 fi
 
 [ "$canonical_status" -eq 0 ] && allow
-[ -n "$canonical_output" ] || canonical_output="Blocked: arming auto-merge on a PR whose reviewDecision is CHANGES_REQUESTED."
+[ -n "$canonical_output" ] || canonical_output="Blocked: this command would arm auto-merge on a PR that cannot merge, or move a failing PR onto a base with no required checks."
 deny "$canonical_output"