@iceinvein/agent-skills 0.15.1 → 0.16.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@iceinvein/agent-skills",
-  "version": "0.15.1",
+  "version": "0.16.0",
   "description": "Install agent skills into AI coding tools",
   "author": "iceinvein",
   "license": "MIT",

package/skills/index.json

@@ -283,7 +283,7 @@
     "name": "sluice",
     "description": "Routes work by change shape into four channels (bypass, fast, main, deep) and applies only the rules each channel needs, so a one-line fix does not pay the cost of a multi-subsystem build. Carries seven rules as one-liners in the router and the full treatment in references read only on friction. Checks the finished plan with plan.sh validate rather than trusting it to memory, seeds the run state from it, keeps a deep run's task breakdown in .sluice/run.json so a statusline segment, one status command and a SessionStart hook can answer where the run is (the hook prints a live run at every session start, compaction included), and closes each run with a ledger read out of the session transcript: elapsed, tools, tokens, and what each dispatched agent cost where the transcript recorded it. Claude Code only; stands down where the superpowers pipeline governs the repo.",
     "type": "prompt",
-    "version": "0.16.1"
+    "version": "0.17.0"
   },
   {
     "name": "temporal-coupling-detector",

package/skills/sluice/SKILL.md

@@ -155,7 +155,9 @@ there at all, which `references/deep-channel.md` handles separately. Review that
 turns out to be missing is only actionable while the plan can still change.
 
 All three answers go into `run.json` and the run record before Task 1's first
-edit, the answer in the first and the reason in the second. Those rows are what
+edit, the answer in the first and the reason in the second, and both files
+open inside the worktree once it is cut, not in the main tree before it
+exists. Those rows are what
 discharge pre-flight, not the approval: one reply arrives for several
 obligations, so a "yes" with no rows behind it signed off the plan and nothing
 else.
@@ -167,9 +169,9 @@ something, not quietly ship without one, and the run ends: `status.sh final`
 when the plan's own review clears, `status.sh close` once the work is no longer
 yours to act on, after a local merge, when the branch is left as it stands, or
 when an open PR lands, not when it opens. A run left open reads as live to the
-statusline and blocks the next one. The SessionStart hook prints a live run at every session start,
-compaction and resume included, so a run you did not start is one you were
-shown, not one you have to remember.
+statusline and blocks the next one. The SessionStart hook prints a live run at
+every session start, compaction and resume included, so a run you did not
+start is one you were shown, not one you have to remember.
 
 ## Conflicts
 

package/skills/sluice/references/deep-channel.md

@@ -89,9 +89,14 @@ in follows from who has to read it.
 and commit, its tier, its `Model` mark and its `Flips` line, and the answers
 pre-flight settled. `scripts/status.sh` writes and reads it, and
 `references/status.md` carries the commands and the statusline segment that
-makes a run visible without anyone asking. Open it with `init` when you open the
-record, seed the rows with `scripts/plan.sh import <plan>` rather than typing a
-command per task, then flip each task as it moves. It carries the ids, the names,
+makes a run visible without anyone asking. Open it with `init` when you open
+the record, which is after pre-flight and inside the tree the work runs in, the
+worktree when pre-flight bought one: a run opened in the main tree before the
+worktree exists lives where every later session's `init` collides with it, and
+that session cannot tell an abandoned run from yours. One already stranded
+there moves with `status.sh move --to <worktree>`. Seed the rows with
+`scripts/plan.sh import <plan>` rather than typing a command per task, then
+flip each task as it moves. It carries the ids, the names,
 the flip, the model marks and the tiers, the last of these floored off `Touches`
 and the contract graph rather than guessed. Import is safe to re-run: a status, a
 review mark or a ratified model already recorded is left alone and a tier is only
@@ -144,10 +149,11 @@ mode has exited and the plan is written. One enforced gate does not collapse two
 stops into one; it only makes the first of them hold.
 
 **The harness's plan file is not the artifact.** It belongs to the mode and not
-to the run. On approval, write the design to `docs/specs/YYYY-MM-DD-<topic>.md`,
-the plan to `docs/plans/YYYY-MM-DD-<topic>.md`, and open the run record and
-`run.json`. Those are the durable files, the ones a session resuming next week
-reads, and none of them is the one you drafted in.
+to the run. On approval, write the design to `docs/specs/YYYY-MM-DD-<topic>.md`
+and the plan to `docs/plans/YYYY-MM-DD-<topic>.md`. The run record and
+`run.json` open after pre-flight, in the tree the work runs in, for the reason
+the run record section gives. Those are the durable files, the ones a session
+resuming next week reads, and none of them is the one you drafted in.
 
 Where plan mode is unavailable, the prose stop is what you have and it is the
 same stop: end the turn on the design and let the next instruction start the
@@ -215,9 +221,27 @@ is the declared schedule the dispatch rules reject.
 If one of the two has only one live answer, say which and ask the other. A stop
 down to a single question is still a stop.
 
-**Write the answers down before Task 1's first edit.** Both files:
-`status.sh preflight` for the answers, the run record for the reason each one
-went that way. That pair is what discharges pre-flight, rather than the
+**Write the answers down before Task 1's first edit, in the tree the work runs
+in.** The order on the instruction that follows the stop: cut the worktree
+first, when that was the answer, through the harness's worktree tool. A fresh
+worktree branches from the remote's default branch, so nothing uncommitted or
+unpushed in the main tree comes across, and `docs/` may not exist there yet.
+Move the design and plan into it yourself, each into its own directory since
+the two share a basename and one `mv` into one directory would leave the plan
+where the design was:
+
+```
+mkdir -p <worktree>/docs/specs <worktree>/docs/plans
+mv docs/specs/YYYY-MM-DD-<topic>.md <worktree>/docs/specs/
+mv docs/plans/YYYY-MM-DD-<topic>.md <worktree>/docs/plans/
+```
+
+The main tree is then clean and the worktree holds the only copy. Then `init`,
+`plan.sh import` and `status.sh preflight` from inside it; then the record
+with the reason each answer went that way; then commit design, plan and record
+there, on the branch the work is on rather than on master. Both files carry the
+answers: `status.sh preflight` for what was decided, the run record for why.
+That pair is what discharges pre-flight, rather than the
 approval you got, and the distinction is the whole point: a stop
 that carries the plan and pre-flight together has one reply for two obligations,
 so a bare "yes" satisfies the plan and leaves no trace either way of the
@@ -293,9 +317,11 @@ reads the run state rather than the plan: it sees what has actually landed.
 - Isolate the workspace before a multi-task plan: the harness's worktree
   tool, not `git worktree` yourself. Implementing straight onto main or
   master needs your partner's say-so, which pre-flight is where you got, and
-  it forecloses concurrent implementers for the whole run. The run state
-  follows the set rather than the tree, so a worktree cut after the plan still
-  reads the rows the plan seeded and flips them where you are watching.
+  it forecloses concurrent implementers for the whole run. Cut it before the
+  run opens, so the state lives in it. A worktree cut after `init` still reads
+  the main tree's run, so nothing breaks for you, but the run stays in the main
+  tree where the next session to start a `deep` run finds it blocking `init`;
+  `status.sh move --to <worktree>` puts it where it belongs.
 - **The agent that built the task commits it**, once its own tests pass, and
   only the paths in its `Touches`. Never `git add -A`: the tree is shared, and
   on a branch you did not isolate it holds work that is not this task's. The

package/skills/sluice/references/status.md

@@ -21,13 +21,16 @@ bash <skill-dir>/scripts/status.sh preflight --review "tier 3 only" --model "6 o
 bash <skill-dir>/scripts/status.sh show
 bash <skill-dir>/scripts/status.sh ready
 bash <skill-dir>/scripts/status.sh final
+bash <skill-dir>/scripts/status.sh move --to <worktree>
 bash <skill-dir>/scripts/status.sh line --full
 bash <skill-dir>/scripts/status.sh close
 ```
 
-`--dir <path>` reads another tree, which is what the statusline uses; it names
-a tree in the set rather than a state file, so a worktree resolves to the same
-run as the tree it was cut from. Statuses
+`--dir <path>` reads another tree, which is what the statusline uses. A tree
+with a run of its own is read as itself; one with none resolves to the main
+worktree of its set, which is the tree a worktree is cut from and not the
+controller's own worktree, so once the run lives there `--dir <implementer
+worktree>` finds nothing. Statuses
 are `todo`, `active`, `review`, `done` and `blocked`. A new id needs `--name`;
 after that every call is a bare flip, so keeping it current costs one command
 per transition rather than a paragraph. `close` archives the run under
@@ -44,8 +47,16 @@ which is the row the debt count counts.
 
 A task going `active` with no `--base` takes the HEAD of the tree the command
 is pointed at, `--dir` if given and the current tree otherwise, once; a base
-already on the row is kept. Pass `--base` when the implementer's tree is
-neither.
+already on the row is kept. Issued from the controller's tree that is the
+controller's HEAD, which is what an implementer worktree cut from that branch
+starts at, so the default is right at dispatch. Where the implementer's tree
+has moved on, pass `--base $(git -C <implementer worktree> rev-parse --short
+HEAD)` rather than `--dir` that tree: with the run in your worktree, `--dir`
+pointed at the implementer's resolves to the main tree and finds no run.
+
+`init` reports any other run live in a tree of the same set, without refusing:
+two sessions in two worktrees is legal, and a run stranded in the main tree
+beside a fresh one in a worktree looks the same until someone says so.
 
 Every write stamps `updated`. Past a day since the last one, `show` and the
 statusline both say how long the run has sat idle, because a finished plan
@@ -69,37 +80,35 @@ that `.gitignore` if you want a run tracked; it is only written when absent.
 
 ## Worktrees
 
-Ignoring itself is what makes the run invisible to a worktree unless something
-is done about it, and a `deep` run makes worktrees after the plan is written:
-`git worktree add` gives the implementer a clean checkout, and an ignored
-directory is not in a checkout. Read from the tree it was called in, the run
-the plan seeded would be absent from every implementer, `init` there would
-start a second run nothing else reads, and the worktree would take that state
-with it when it went.
-
-So a tree with no run of its own anchors on the main worktree of its set, and
-the set's run covers it: the statusline renders the controller's run in every
-implementer's window, and a flip issued from any of those trees lands where
-every other one is watching. A tree's own run comes first, though. `init`
-always lands in the tree it is given, and every other command reads that
-tree's state when it has one, so two sessions working independently in two
-worktrees of one repo each keep their own run and neither is shown the
-other's. Anchored unconditionally, as this once was, the first `init` in the
-set took over every other session's statusline and refused every other `init`.
-A submodule anchors on its own checkout, not the superproject's, and a
-directory that is no git work tree keeps its run exactly where it sits.
+The run lives with the controller: in the worktree the work runs in when
+pre-flight bought one, in the main tree otherwise. Implementers never read or
+write it, the controller flips every row, so nothing an implementer does
+depends on seeing the run from its own worktree, and two sessions working
+independently in two worktrees of one repo each keep their own run with
+neither shown the other's.
+
+`init` therefore always lands in the tree it is given, and every other command
+reads that tree's own state when it has one. A tree with none falls back to the
+main worktree of its set, which is what keeps a session working before this
+rule existed, run in the main tree and worktree cut afterwards, reading the run
+it started. That run stays in the main tree, where it blocks the next
+session's `init`; `move --to <worktree>` relocates it, refusing a destination
+that already holds a run or that is not a work tree of the same repository. A
+submodule anchors on its own checkout, not the superproject's, and a directory
+that is no git work tree keeps its run exactly where it sits.
 
 One file for several writers is one file to contend on, so `init`, `task`,
-`preflight`, `final` and `close` take a lock first: two flips issued at the
-same moment from different trees would otherwise have the later write built on
-a snapshot taken before the earlier one landed, dropping that row without
-saying so. The lock
+`preflight`, `final`, `close` and `move` take a lock first, `move` taking the
+destination tree's as well as its own: two flips issued at the same moment
+from different trees would otherwise have the later write built on a snapshot
+taken before the earlier one landed, dropping that row without saying so. The lock
 carries its holder's pid, so a killed run is broken through rather than waited
 out. Reads take nothing, state being installed through a rename, which is what
 keeps `line` cheap enough to render on.
 
 Open it with `init` when you open the run record, at the same point and for the
-same reason, then seed the rows with `plan.sh import <plan>` rather than a
+same reason, and in the same tree: after pre-flight, inside the worktree when
+one was bought. Then seed the rows with `plan.sh import <plan>` rather than a
 command per task. The ids, names, the flip, the `Model` marks and the tiers are
 all fixed the moment the plan is written and are already in the file, so typing
 them again is transcription with a chance of error in it.
@@ -263,11 +272,12 @@ then print it last, after whatever else the command emits:
 if [ -n "$sluice_line" ]; then printf '%s\n' "$sluice_line"; fi
 ```
 
-The gate is two tests because a linked worktree holds no state file of its own:
-there `.git` is a regular file naming the tree it was cut from, and the script
-resolves the run from it. In a tree with no run and no worktree behind it `.git`
-is a directory, so both tests fail and no process is spawned, which is the
-property the gate is for.
+The gate is two tests because a linked worktree may hold no state file of its
+own and still belong to a set with a run: there `.git` is a regular file naming
+the tree it was cut from, and the script falls back to that tree. A worktree
+with its own run passes the first test. In a tree with no run and no worktree
+behind it `.git` is a directory, so both tests fail and no process is spawned,
+which is the property the gate is for.
 
 `if` rather than `[ ... ] &&`: as the last command of a statusline script the
 short form makes it exit 1 on every render with no run live, which is the common

package/skills/sluice/scripts/status.sh

@@ -16,6 +16,7 @@
 #   status.sh show [--json]
 #   status.sh ready
 #   status.sh final
+#   status.sh move --to <tree>
 #   status.sh line [--full]
 #   status.sh close
 #
@@ -24,7 +25,8 @@
 # directory ignores itself, so no project needs a .gitignore line for it.
 #
 # Exit: 0 ok, 1 the state could not be written, 2 no live run, 3 a run is
-# already live, 4 bad arguments, 5 jq missing, 6 the state file is unreadable.
+# already live (here, or at move's destination), 4 bad arguments, 5 jq missing,
+# 6 the state file is unreadable.
 # `line` is exempt and always exits 0 in silence, because a statusline renders
 # on every keystroke and has nowhere to put an error.
 #
@@ -40,7 +42,7 @@ err() { echo "status.sh: $*" >&2; }
 
 usage() {
 	echo "usage:" >&2
-	sed -n '/^#   status.sh init/,/^# 5 jq missing/p' "$0" | sed 's/^# \{0,2\}//' >&2
+	sed -n '/^#   status.sh init/,/unreadable\.$/p' "$0" | sed 's/^# \{0,2\}//' >&2
 }
 
 # A flag's value has to be checked before `shift 2`, not after. Bash refuses to
@@ -269,9 +271,9 @@ require_readable() {
 # runs in having to add a line to its own .gitignore. `*` matches the .gitignore
 # file too, so the whole directory drops out of `git status`. An existing file is
 # left alone, and a tree that refuses the write still gets its run.
-mk_dir() { # <directory to create under .sluice>
+mk_dir() { # <directory to create under .sluice> [<tree whose .sluice it is, default $DIR>]
 	mkdir -p "$1" || { err "could not create $1"; exit 1; }
-	local ignore="$DIR/.sluice/.gitignore"
+	local ignore="${2:-$DIR}/.sluice/.gitignore"
 	[ -e "$ignore" ] || printf '*\n' >"$ignore" 2>/dev/null || true
 }
 
@@ -313,30 +315,31 @@ write_state() {
 # later one. A lock whose holder is gone is broken rather than waited out, and
 # one whose holder is alive is waited on for a bounded time and then reported,
 # because a command that hangs in a status bar is worse than one that fails.
-lock_taken=0
+# `move` spans two trees and holds both locks, so the locks taken are a list.
+LOCKS_TAKEN=()
 release_lock() {
-	[ "$lock_taken" -eq 1 ] || return 0
-	rm -rf "$LOCK"
-	lock_taken=0
+	local l
+	for l in "${LOCKS_TAKEN[@]+"${LOCKS_TAKEN[@]}"}"; do rm -rf "$l"; done
+	LOCKS_TAKEN=()
 }
 
-take_lock() {
-	local waited=0 holder
-	while ! mkdir "$LOCK" 2>/dev/null; do
-		holder="$(cat "$LOCK/pid" 2>/dev/null)"
+take_lock() { # [<lock path>, default the run's own]
+	local lock="${1:-$LOCK}" waited=0 holder
+	while ! mkdir "$lock" 2>/dev/null; do
+		holder="$(cat "$lock/pid" 2>/dev/null)"
 		if [ -n "$holder" ] && ! kill -0 "$holder" 2>/dev/null; then
-			rm -rf "$LOCK"
+			rm -rf "$lock"
 			continue
 		fi
 		if [ "$waited" -ge 100 ]; then
-			err "another sluice command has held $LOCK for 10s; remove it if nothing is running"
+			err "another sluice command has held $lock for 10s; remove it if nothing is running"
 			exit 1
 		fi
 		sleep 0.1
 		waited=$((waited + 1))
 	done
-	printf '%s\n' "$$" >"$LOCK/pid" 2>/dev/null || true
-	lock_taken=1
+	printf '%s\n' "$$" >"$lock/pid" 2>/dev/null || true
+	LOCKS_TAKEN+=("$lock")
 	trap release_lock EXIT INT TERM
 }
 
@@ -361,10 +364,25 @@ case "$SUB" in
 		take_lock
 		if [ -f "$STATE" ] && [ "$FORCE" -eq 0 ]; then
 			live="$(jq -r '.topic // "?"' "$STATE" 2>/dev/null || echo "?")"
-			err "a run is already live (topic: $live); pass --force to replace it"
+			err "a run is already live (topic: $live). If it is yours and your work runs in a worktree, put it there: status.sh move --to <worktree>. If it is another session's, leave it and start yours from your own worktree. If it is finished, status.sh close; --force replaces it"
 			exit 3
 		fi
 
+		# Other trees in the set may hold runs of their own, legitimately or as
+		# one stranded before its session moved into a worktree. Said, never
+		# refused: the session starting here cannot tell which, and the one that
+		# can is the owner.
+		if [ "$(git -C "$DIR" rev-parse --is-inside-work-tree 2>/dev/null)" = "true" ]; then
+			here="$(cd "$DIR" && pwd -P)"
+			git -C "$DIR" worktree list --porcelain 2>/dev/null | sed -n 's/^worktree //p' | while IFS= read -r tree; do
+				[ -d "$tree" ] || continue
+				[ "$(cd "$tree" && pwd -P)" != "$here" ] || continue
+				[ -f "$tree/.sluice/run.json" ] || continue
+				other="$(jq -r '.topic // "?"' "$tree/.sluice/run.json" 2>/dev/null || echo "?")"
+				err "note: another run is live in $tree (topic: $other); if it is this work stranded before a worktree was cut, move it there instead"
+			done
+		fi
+
 		jq -n \
 			--arg topic "$TOPIC" \
 			--arg channel "$CHANNEL" \
@@ -664,6 +682,53 @@ case "$SUB" in
 		jq --arg now "$(date -u +%Y-%m-%dT%H:%M:%SZ)" '.final_review = $now' "$STATE" | write_state
 		;;
 
+	move)
+		TO=""
+		while [ $# -gt 0 ]; do
+			case "$1" in
+				--to) need_value --to $# "${2-}"; TO="$2"; shift 2 ;;
+				*) err "unknown flag: $1"; exit 4 ;;
+			esac
+		done
+		[ -n "$TO" ] || { err "move needs --to <tree>"; exit 4; }
+		[ -d "$TO" ] || { err "no such directory: $TO"; exit 4; }
+		TO="$(cd "$TO" && pwd -P)"
+		require_run
+
+		# The destination is taken as given rather than anchored: the whole point
+		# is to put the run in one particular tree, the controller's worktree,
+		# which anchoring would resolve straight back to the main tree it is
+		# leaving. It does have to be a work tree of the same set: a typo would
+		# otherwise strand the run at a path no command issued from the tree
+		# resolves, and the only way back is knowing where it went.
+		src_common="$(git -C "$DIR" rev-parse --git-common-dir 2>/dev/null)"
+		dst_common="$(git -C "$TO" rev-parse --git-common-dir 2>/dev/null)"
+		dst_top="$(git -C "$TO" rev-parse --show-toplevel 2>/dev/null)"
+		if [ -n "$src_common" ]; then
+			src_common="$(cd "$DIR" && cd "$src_common" 2>/dev/null && pwd -P)"
+			dst_common="$([ -n "$dst_common" ] && cd "$TO" && cd "$dst_common" 2>/dev/null && pwd -P)"
+			if [ -z "$dst_common" ] || [ "$dst_common" != "$src_common" ] || [ "$dst_top" != "$TO" ]; then
+				err "$TO is not a work tree of the same repository as $(cd "$DIR" && pwd -P); move only relocates a run between trees of one set"
+				exit 4
+			fi
+		fi
+		[ "$TO" != "$(cd "$DIR" && pwd -P)" ] || { err "the run is already in $TO"; exit 4; }
+
+		# Both trees' locks: a racing init in the destination takes that tree's
+		# lock, not this one's, and the mv would land on top of what it wrote.
+		take_lock
+		mk_dir "$TO/.sluice" "$TO"
+		take_lock "$TO/.sluice/run.lock"
+		DEST_STATE="$TO/.sluice/run.json"
+		if [ -f "$DEST_STATE" ]; then
+			there="$(jq -r '.topic // "?"' "$DEST_STATE" 2>/dev/null || echo "?")"
+			err "a run is already live in $TO (topic: $there); close it there first"
+			exit 3
+		fi
+		mv "$STATE" "$DEST_STATE" || { err "could not move $STATE to $DEST_STATE"; exit 1; }
+		echo "moved $(jq -r '.topic // "run"' "$DEST_STATE" 2>/dev/null || echo run) to $TO"
+		;;
+
 	close)
 		[ $# -eq 0 ] || { err "close takes no arguments"; exit 4; }
 		require_run

package/skills/sluice/skill.json

@@ -1,6 +1,6 @@
 {
   "name": "sluice",
-  "version": "0.16.1",
+  "version": "0.17.0",
   "description": "Routes work by change shape into four channels (bypass, fast, main, deep) and applies only the rules each channel needs, so a one-line fix does not pay the cost of a multi-subsystem build. Carries seven rules as one-liners in the router and the full treatment in references read only on friction. Checks the finished plan with plan.sh validate rather than trusting it to memory, seeds the run state from it, keeps a deep run's task breakdown in .sluice/run.json so a statusline segment, one status command and a SessionStart hook can answer where the run is (the hook prints a live run at every session start, compaction included), and closes each run with a ledger read out of the session transcript: elapsed, tools, tokens, and what each dispatched agent cost where the transcript recorded it. Claude Code only; stands down where the superpowers pipeline governs the repo.",
   "author": "iceinvein",
   "type": "prompt",