Herdr Standup Skill

Process steps in order. Do not skip ahead.

A standup is not a round of work: nothing is dispatched, no context is cleared, and no worker is interrupted. A worker that is mid-task keeps working and its row comes from what you already know.

Each command block resolves CP to the project-local plugin, falling back to $HOME/.tessl/plugins/jbaruch/coding-policy. Run the resolver in every call. Prose skills/... paths are relative to that plugin root.

Step 1 — Roster the Team

CP=.tessl/plugins/jbaruch/coding-policy; [ -d "$CP" ] || CP="$HOME/$CP"
bash "$CP/skills/herdr-teamlead/roster.sh"

Emits {"caller":{...},"agents":[{"name","kind","pane_id","state"}]} for every named agent other than your own pane. Exit 1 is a precondition (outside Herdr, herdr absent); exit 2 is a herdr failure. On either, report the message verbatim and finish here.

Split the roster by state:

• idle or done — ask them in Step 2.
• working or blocked — never asked. Their row comes from the round log or the assignment ledger, with what they are on, in Step 4.

An empty roster means there is no team to stand up. Say so and finish here. Proceed immediately to Step 2.

Step 2 — Ask Each Ready Worker

One call per idle or done worker:

CP=.tessl/plugins/jbaruch/coding-policy; [ -d "$CP" ] || CP="$HOME/$CP"
bash "$CP/skills/herdr-standup/standup-ask.sh" \
  <agent-name> <absolute-report-path>

Sends the standup question as a plain message and emits {"agent","report_path","state","sent"}. Exit 3 means the worker was not ready and nothing was sent — move it to Step 4's list. Exit 1 is a precondition, including a report path longer than the script's limit (the worker's REPORT: <path> line must fit one pane row), exit 2 a herdr failure. The question text and the four-line shape it demands are the script's contract; see the header of skills/herdr-standup/standup-ask.sh.

Proceed immediately to Step 3.

Step 3 — Wait for Each Answer

Wait for each worker asked in Step 2:

CP=.tessl/plugins/jbaruch/coding-policy; [ -d "$CP" ] || CP="$HOME/$CP"
bash "$CP/skills/herdr-standup/standup-wait.sh" \
  <agent-name> <absolute-report-path>

Same argv, stdout, and exit codes as wait-report.sh. The budget is the script's constant, never a number chosen here; see the header of skills/herdr-standup/standup-wait.sh. Exit 1 means the worker did not answer inside it. It is not chased twice. Move it to Step 4's list with what you know. Exit 3 means a dialog is up — relay it and leave that worker to the operator. Exit 5 means the report is unavailable after a confirmed provider refusal. Record it as missing in Step 4 and notify the operator; never automatically retry, rephrase, switch models/providers, or synthesize an answer. Exit 4 means the answer file exists but the pane did not show the marker whole. That is not an answer. Read the live state with herdr agent get <agent-name> and take the first continuation that applies:

• The command fails — report its message verbatim and finish here.
• The state is blocked or working — re-run the wait for that worker once. Other exits take their documented branches. A second exit 4 is terminal: move the worker to Step 4's list with what you know.
• The state is idle or done — move the worker to Step 4's list with what you know. standup-ask.sh refuses a report path too long for one pane row, so this outcome means the path bypassed it.

Proceed immediately to Step 4.

Step 4 — Write the Rows Nobody Answered

For every worker that was busy, blocked, or silent, write one entry into a JSON file for the renderer:

{
  "<agent>": {
    "roles": "developer",
    "done": "<from the round log>",
    "plan": "<from the round log>",
    "blocked": "none",
    "note": "busy: <task>"
  }
}

note renders beside the agent's name, so a busy worker reads as grok (busy: refactor). Every field is optional. Take the content from the round log or the assignment ledger — never from a pane read, and never from a guess about what a worker is probably doing. Use the lead's schema-1 task ledger under skills/herdr-teamlead/state-schema.md for accepted completion. Read it without writing or migrating it. Missing or unsupported records mean unknown completion; dispatch status alone establishes none. Worker DONE answers remain self-reports.

If every worker answered, skip the file. Proceed immediately to Step 5.

Step 5 — Render the Standup

CP=.tessl/plugins/jbaruch/coding-policy; [ -d "$CP" ] || CP="$HOME/$CP"
python3 "$CP/skills/herdr-standup/standup-render.py" \
  --reports <round-reports-dir> \
  --now <ISO-8601> \
  --team "<label>" \
  --agent <name>=<report-path> [--agent ...] \
  [--roles <roles-json>] [--extra <extra-json>]

Writes standup-<date>.md into the reports directory and prints one JSON object on stdout: {"markdown_path","block","answered","unasked"}, where block is the fenced terminal table. Save the output to a file. Exit 2 means a report file did not carry the four-line shape, naming the file — ask that worker again, or move it to --extra. Exit 1 is a missing or unwritable input, and the message says what to fix.

--now is required and never defaults to the clock, so the same inputs render the same bytes. The column widths and the wrapping are the script's contract; see the constants at the top of skills/herdr-standup/standup-render.py.

Proceed immediately to Step 6.

Step 6 — Relay the Block

Read saved attention using the team's recorded state override or default:

CP=.tessl/plugins/jbaruch/coding-policy; [ -d "$CP" ] || CP="$HOME/$CP"
bash "$CP/skills/herdr-teamlead/teamlead.sh" catch-up [--state <state-file>]

Present nonempty attention_markdown before the table. Follow attention page counts and next_offset before claiming every pending item was shown. The read never resolves an item, updates a presentation marker, or contacts workers. On non-zero, report that saved attention is unavailable and its diagnostic; continue with the independently gathered standup. If there are no pending items, proceed silently. Its lifecycle contract is:

skills/herdr-teamlead/references/attention.md

Print block from the saved output verbatim (jq -r .block <file>), exactly as the renderer emitted it. Do not reformat it, summarize it, or replace it with prose.

Name markdown_path underneath, in one line. Add a provenance line identifying worker-reported rows and rows whose completion was verified from the task ledger. Label unavailable acceptance evidence as unknown; dispatch records prove none. Add your own reading of the standup only if a row carries something the operator should act on today — a blocker naming another worker, or a plan that contradicts the round in flight. Otherwise the table speaks for itself. Finish here.