watchbill

Plugin watchbill from marketplace watchbill in johntrandall/claude-plugin-watchbill. Commands read from the repo's manifest — 26 days ago.

Skills (15)

• architect

  Obligations of the PLAYBOOK ARCHITECT seat (the architect): run the build loop, brief the cycle and its rules leads, edit in a git worktree and land with `watchbill land` between the steward's items, send the steward FYI announcements (no ack wait), record build events as Watchbill-Event commit trai

• collaborative-drafting

  Drafting a document with the human through an EDITOR seat: one editor owns the text and proposes every change as a signed Track Changes suggestion with a reason; every other agent only comments; the human accepts or rejects; the editor accepts the human's comment-free suggestions and never rejects t

• delegation

  How a watchbill seat launches, directs and closes a delegate (crew sub-agent, lead in its own tab, teammate): pick one of four presets over the delegation properties, use the watchbill plugin agents that pin each model, put the launcher of record's authority in the brief, record lineage (launched_by

• handoff-session

  Starting a successor or an independent Claude session in its own herdr tab: when to offer a handoff (85 % context, or the cost band), releasing what only you hold, the briefing, spawning with herdr-lifecycle, the fork variant that carries your history, verifying the spawn, relabelling and announcing

• herdr-workspace-placement

  Which herdr workspace (shelf) a session's tab belongs in once its work changes shape, and the order of tabs in a workspace: a session that only monitors shelves itself on the monitoring shelf with the 📡 glyph (`herdr-lifecycle shelve monitoring`), one that only waits on something other than the hum

• historian

  How a handed-off origin (tab ↪) behaves and ends: stay idle in the project workspace answering history questions, redirect new work to the head of the chain, never poll, exit only when herdr-lifecycle historian-check allows it or the human says so, the exit steps, the proxy historian for an origin o

• infra-handoff

  Handing a fault outside your own subject (an infrastructure problem: a lost authorization, a failing service or token feeder, a network or storage fault, a credential to re-provision) to its own workspace and orchestrator, in the subject's home, then stepping back. Auto-invoke when a session working

• lead-and-crew

  How a lead runs a multi-step change with a crew: a task list with dependencies, phases of parallel crew in worktrees (watchbill crew sub-agents, or herdr worktree workers of any agent kind), merging and verifying between phases, cleaning up worktrees, a final independent verifier, and the recovery c

• monitoring-mode

  How a session behaves once its work is delivered and only a watch keeps it alive: write the watch contract (probe, exit condition, deadline, alert or note, cadence), pick the mechanism, stay silent on no-op ticks, back off, escalate once, stop polling what only a person can clear, check your own con

• multi-agent-routing

  Choosing how to get more than one Claude working (sub-agent, fork sub-agent, agent team, cross-session message, handoff or fork session, scheduled run, workflow) by who the helper reports to, and the canonical rules for messaging a running session: SendMessage not keystrokes, why (authority), what s

• paper-trails

  Who records what, where, and when in a watchbill system: the seat that holds the why writes it at action time, tools stamp clocks and counts, build events are Watchbill-Event commit trailers, every item carries the rules SHA it ran under, every decision of the human is recorded where its item lives

• role-relaunch

  Replacing a running role session (one with an agent definition) with a fresh one started from that definition: when (its definition or hooks changed, it was started without --agent or -n, or its context is full outside a watchbill relief), the handoff note and kickoff kept in the project, spawning w

• steward

  Contract for the STEWARD seat: the standing agent that runs a system's inbox (new items, the human's decisions, confirmations) in the primary checkout under the current rules, stamps the rules SHA on every item and adopts landed rules only between items after its check passes, never edits the rules,

• watchbill

  The role model for a multi-agent system run in herdr with a human in the loop: which seat you are (architect, steward, rules or run lead, standing lead, editor, crew, advisor, historian), the three write classes (rules, work, nothing), the plugin agents that pin each seat's model, how seats message

• watchers

  Writing or choosing a watcher (what counts as a change, exit codes, seen-state): a watcher detects change and the woken seat acts; exact change definition on structured fields; settle on quiet; seen-state outside the watched tree and /tmp; bounded life with re-arm; exit codes 0 change, 1 timeout, 2

Subagents (12)

• advisor

  watchbill ADVISOR seat: answers design questions on call from its history or reading; changes nothing. Use for an on-call advisor tab (for example a resumed design session) or a one-off design question.

• architect

  watchbill PLAYBOOK ARCHITECT seat: changes a system's playbook (its rules: schema, code, skills, briefs, README) in its own git worktree, lands with `watchbill land` between the steward's items (Watchbill-Event trailers, single-use Watchbill-Approved), writes the gate and watch.md, parks with 👁. Us

• backfill

  watchbill crew BACKFILL: applies one landed rule change to existing records, in one commit with a Watchbill-Event: backfilled trailer. Use when an architect lands a rule that existing records must catch up with.

• checker

  watchbill crew CHECKER / commenter: reviews a draft and leaves comments only, through the review tool; never edits text. Use when a draft needs a facet review (style, facts, layout) beside the editor.

• editor

  watchbill EDITOR seat: the standing run lead that owns one document's text with the human, proposing every change as a signed Track Changes suggestion and fielding the human's marks with a watcher. Use to launch the editor of a document.

• escalation

  watchbill crew ESCALATION specialist: takes one item a Sonnet worker failed and does it, or says exactly why it needs the human. Use after a worker fails an item, before marking it needs-human.

• reader

  watchbill crew READER / researcher: reads, searches and reports; writes nothing. Use from any seat for a search or reading sweep whose file dumps should stay out of the seat's context.

• rules-lead

  watchbill RULES LEAD seat (task or standing): changes rules only inside the files its brief assigns, runs its own crew, reports to the architect that spawned it. Use when an architect launches a lead for one bounded rules task or a standing rules function.

• run-lead

  watchbill RUN LEAD seat (task or standing): does a chunk of a system's work under the current rules at a stated rules SHA, never edits rules, sends rule problems to the steward. Use when a steward launches a lead for a batch or a standing work function.

• steward

  watchbill STEWARD seat: runs a system's inbox (new items, the human's decisions, confirmations) under the current rules, stamps the rules SHA on each item, records every outcome with its reason, never edits rules. Use to launch or relaunch a system's steward (claude --agent watchbill:steward).

• verifier

  watchbill crew VERIFIER: checks a change or an item cold and reports pass or fail per check, with script evidence; writes nothing. Use after a rules lead lands a change or before a gate closes.

• worker

  watchbill crew WORKER: one step of real work, writing only its records through the system's tools, then a report. Use from any seat for a step that would add more than about 20K tokens to the seat's context.

Hooks (5)

• SessionStart

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/session_context.py"

• UserPromptSubmit

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/context_band.py"

• PreToolUse

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/hold_guard.py"

• PreToolUse

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/ownership_warn.py"

• PreToolUse

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/model_guard.py"

Similar plugins