A report-only template for a daily standup over your open issues and scheduled agents. It reports what moved, what has gone quiet, what looks finished but was never closed, and which agents ran. It never comments, reassigns, or changes a status.
↓ Download SKILL.mdmkdir -p .claude/skills/program-overseer
cp SKILL.md .claude/skills/program-overseer/SKILL.mdThe skill ships as a template. Every value that depends on your setup is a placeholder, including the definitions of "gone quiet" and "finished". You supply those definitions; the skill does not assume them.
Issues stop moving without any event that announces it. The board keeps showing them as open and in progress.
Completed tasks that never get closed make the board overstate what remains, and readers stop trusting it.
A scheduled agent that stops running produces no output, so nothing draws attention to it.
Checking each of these by hand repeats every day and grows with every agent and program added.
The skill lists the actions it will not take: commenting, assigning, changing status, creating or closing items, and editing scheduled agents. When a step seems to need one of them, the finding goes in the report and the skill stops.
Compliance, risk, exception, engineering, and operations work all use the same report format. The skill reports the state of an item and does not interpret its substance.
When the tracker or agent log cannot be read, the report says so under Notes. The skill does not fill a gap with a guess.
Ask for the standup manually each morning for a while. Read the report against your own sense of the work and note what feels wrong.
Change your definitions of quiet and finished before changing the report format. Most early mismatches come from definitions.
Attach the skill to a scheduled task at a fixed time. A fixed time makes a missing report noticeable.
Add a section only after you have performed the same manual check twice. Remove sections that stop changing decisions.
A definition that fits another team's workflow produces noise in yours. Build definitions from your own items.
Write access puts the agent's mistakes into the record and creates a second review job. The report-only constraint keeps the failure cost at one minute of reading.
A long first report gets skimmed. Start with four groups and let each addition earn its place.
The report describes state. It does not assign work. Decisions about each item stay with the reader.
Definitions of quiet and finished depend on how your team uses its tracker. Borrowed definitions generate noise.
Workflows change. A definition that matched practice last quarter can stop matching without any visible sign.
A dashboard shows current state and waits for a visit. This skill produces a report on a schedule that states what changed, so the reader does not need to remember to check.
Every heading reads "None" and the report still appears. An empty report confirms that the run happened and that the agent-runs heading found nothing wrong.
The template forbids it. A separate agent with its own review step is the place for actions, and that agent needs a different design from this one.
Any tracker Claude can read through a connector or an export. Describe the access method in the TRACKER placeholder.
Let the format settle over several runs before adding checks. A stable format makes changes between days visible.
When you catch yourself performing the same manual check a second time, give it a heading and a definition in the file.
Give the skill a read-only connection to your tracker. The constraint then holds even if the instructions are misread.
A report that arrives at the same time each weekday makes a missing report noticeable.
Runs a daily standup over a body of work and reports the result. It asks what moved, what is stuck, and what is done. The people who own the work keep the work.
The skill reads and reports. It does not comment, assign, change status, create or close items, edit scheduled agents, or message anyone on the reader's behalf.
Placeholders for the tracker, scope, agent log, schedule, report destination, your definition of quiet, your completion signs, and optional categories. The skill asks for any required value that is blank.
Read the configuration, read open items, read the agent log, read the previous report, sort items into groups, record agent runs, write the report, and stop.
Fixed headings: Moved, Gone quiet, Finished still open, Agent runs, and Notes. Unreadable sources go under Notes, and the skill reports only what it could read.
A seven-step method for setting your own definitions of quiet and finished from items you know are healthy, stuck, or done, and for reviewing them as your workflow changes.
A short table of the reader's typical response to each group. The skill does not perform or recommend these responses.
Start with four groups. Add a new check only after the reader has performed the same manual check twice, and keep the report-only constraint in every version.