Skill Fixture

Daily Standup Coach (Skill classifier fixture)

A pure-prose AI skill that helps teams prepare for and run daily standup meetings. This SKILL.md fixture is used by bin/tests/test_classify_project.py as a known-Skill target for the v1.5.3 project-type classifier: SKILL.md at fixture root, no substantial code, prose word count substantially exceeds code line count.

The fixture is a synthetic skill, not a real one in production use; its content is realistic enough to exercise the classifier but should not be mistaken for a deployable skill.

When to use this skill

The skill activates when an operator asks for help running a daily standup meeting. Triggers include phrases like "help me run standup," "review my standup notes," "draft talking points for tomorrow's standup," or "summarize this week's standups." It can also be invoked as part of a larger sprint ceremony orchestration where standup is one of several meetings the operator needs to facilitate.

The skill does not activate for retrospectives, sprint planning, refinement sessions, or other agile ceremonies. Those meetings have different shapes and different anti-patterns and would need their own skills. If an operator asks the skill to handle a non-standup meeting, the skill politely declines and points at the appropriate alternative tool, if one exists, or suggests the operator use a general-purpose meeting facilitator.

What this skill produces

For each invocation the skill produces three artifacts. The first is a concise meeting agenda: who is presenting, in what order, with how much time each. The agenda follows the round-robin pattern that most teams default to, but the skill will reorder if the operator has flagged that a specific blocker should be discussed first. The second artifact is a set of prompts the operator can use to keep the meeting on track: when to redirect a tangent, when to move someone to "parking lot," when to escalate a recurring blocker. The third artifact is a brief written summary of the meeting that can be posted to a shared channel after the meeting concludes.

The artifacts are written incrementally as the skill processes the operator's input. The agenda lands first because most operators want to review it before the meeting; the prompts land second because they're needed during the meeting; the summary template lands last because it's populated post-meeting from the operator's notes.

Phase 1 — Pre-meeting preparation

Phase 1 runs in the time window between the operator deciding to use the skill and the standup meeting starting. The skill reads any context the operator provides — last week's notes, the team's current sprint goals, known blockers — and produces a draft agenda.

Phase 1's output is agenda.md in the operator's working directory. The file has three sections: roster (who is expected to attend, in presentation order), focus areas (sprint goals being discussed this week), and parking lot (topics that came up but should be deferred). Each section is brief — the agenda is meant to fit on one screen so the operator can keep it open during the meeting.

The operator can edit the agenda before the meeting. The skill respects operator edits and uses the edited agenda as the basis for the next phase rather than regenerating from scratch.

If the operator hasn't provided any context, Phase 1 produces a skeleton agenda with placeholders the operator fills in. This is the most common case for first-time use of the skill — the operator hasn't yet trained the skill on the team's specific dynamics.

Phase 2 — In-meeting facilitation

Phase 2 runs during the standup meeting itself. The operator is presumably busy facilitating, so the skill operates in low-touch mode: it watches a shared notes document the operator updates as the meeting progresses, and emits short prompts when it detects patterns that need redirection.

Patterns the skill watches for include: someone has been talking longer than their allotted time; the same blocker has come up across multiple recent standups; a topic is drifting into deep-dive territory and should be moved to a follow-up meeting; the team is reporting all-green on a goal that other signals suggest is at risk.

When the skill detects one of these patterns, it emits a prompt as a comment in the shared notes. The prompt is brief and actionable: not "this discussion seems to be drifting," but rather "consider parking this for a follow-up; @alice and @bob can sync after standup." The operator can ignore the prompt, apply it, or modify it.

Phase 2's output is the cumulative set of prompts emitted during the meeting. They live in meeting_prompts.md and are timestamped so the operator can correlate them with the meeting recording if one exists.

Phase 3 — Post-meeting summary

Phase 3 runs after the standup meeting concludes. The operator typically hands the skill the raw notes from the meeting plus any audio transcript the team's recording tool produced. The skill produces a written summary suitable for posting to the team's shared channel.

The summary has a fixed structure: what the team accomplished since last standup, what the team is working on now, what blockers exist and who owns each. The structure is fixed because variability in summary structure makes it hard for downstream consumers (the team's tooling, the manager's weekly digest) to extract structured data.

Each section of the summary is a bulleted list. Each bullet is a single sentence. Each blocker bullet names an owner and an expected resolution window — the skill prompts the operator if those fields aren't clear from the notes.

Phase 3's output is summary.md, which the operator reviews before posting.

Phase 4 — Pattern recognition across meetings

Phase 4 runs periodically — typically weekly — and looks across the recent history of summaries to identify patterns. A blocker that has appeared in three consecutive standups without resolution. A team member whose updates have been getting shorter and more vague over time. A sprint goal that the team has been reporting progress on but never closing out.

The skill flags these patterns to the operator without prescribing action. The operator typically knows the team better than the skill does and is in a better position to decide whether a flagged pattern warrants intervention. The skill's job is to surface the pattern, not to act on it.

Phase 4's output is patterns.md, an append-only log of patterns the skill has identified. Each entry has the pattern, the evidence (which standups contributed), and a suggested escalation path the operator can take if they want to address the pattern.

Phase 5 — Retrospective input preparation

Phase 5 is invoked at the end of a sprint, typically during the operator's preparation for the team retrospective. The skill bundles the patterns from Phase 4 plus a summary of the sprint's standups into a retrospective input document.

This is not the retrospective itself — the skill explicitly does not run retrospectives, only standups. The output is meant to be one input among many for the operator (or a separate retrospective skill) to consume.

The output is retro_input.md and is structured around the standard retrospective categories: what went well, what didn't, what to try next. The skill populates each category from standup data; the operator and the team add categories the standup data can't speak to.

Anti-patterns the skill avoids

The skill explicitly does not do the following, even when asked: rate individual contributors against each other; produce performance review input; act as a surveillance tool for tracking who is "really working"; generate disciplinary documentation. These uses misread the standup as a status-reporting meeting rather than a coordination ritual, and the skill declines them with a brief explanation pointing at appropriate alternative practices.

The skill also avoids over-instrumenting standups: it does not push for more meetings, longer meetings, or more detailed reporting unless the operator has explicitly requested those changes. The default bias is toward shorter, lower-touch meetings with clear outcomes.

How the skill calibrates

The skill maintains a small per-team calibration state that captures the team's preferences: how strict to be about time-boxing, how aggressive to be in flagging patterns, how much to summarize versus quote raw notes. The calibration is updated when the operator overrides a skill suggestion ("don't flag this kind of thing again") or when the operator explicitly tunes ("be more aggressive about parking-lot suggestions").

Calibration is persistent across invocations. It lives in calibration.json in the operator's working directory and is portable between machines if the operator copies the file.

Extending the skill

Operators can extend the skill by adding domain-specific reference files under references/. Each reference file describes a class of pattern or intervention specific to the operator's context — for instance, a file describing the team's specific definition of "blocked" or a file listing the senior contributors whose updates should always be parked rather than deep-dived during standup.

The skill reads references/ at the start of each invocation and incorp- orates the reference content into its decision-making. Reference files must be markdown; other formats are not supported.

When SKILL.md and a reference file disagree, SKILL.md wins. Reference files are supplementary, not authoritative.