analyze-agent-transcript-prompt-ambition

name	analyze-agent-transcript-prompt-ambition
description	Per-user-source-New-Trigger analyzer (the case formerly known as the Initial Prompt). Flags user-typed Triggers that look like they under-scoped the work — short, narrow, followed quickly by another user-source New Trigger on a related Goal. Pattern suggests the user split work the agent could have one-shotted with a more ambitious prompt. Output feeds the human-prompting recommendation bucket, and may also surface deterministic-trigger opportunities (the ideal end-state for user-source New Triggers).
user-invocable	false

Analyze prompt ambition

Per-Segment ambition check. Runs only on Segments with trigger.kind == "New" && trigger.source == "user" — i.e. user-typed prompts that opened a fresh Goal, the case formerly known as Initial Prompts.

Inputs

• segment: a Segment with trigger.kind == "New" and trigger.source == "user", carrying its meta block (event_range, wall_clock_s, model). The orchestrator hands you the Segment directly — you do not walk raw JSONL.
• next_user_new_segments: the next 1-3 Segments in the same Transcript with the same Trigger shape (kind: New, source: user) — needed to detect the "user split work" pattern. May be empty when this is a root Segment with no following user-source New sibling.
• transcript.json: the OpenTranscripts Transcript document, available to dereference event ids when needed.
• external_context (optional): external-context.json if present — useful for spotting recurring user-triggers that should be deterministic.
• philosophy_prompting: the philosophy-on-prompting reference — grounds what "ambitious" means (a single end-to-end definition of done, not a longer hand-hold) and why the deterministic trigger is the north-star case.

Output

This is the item body. The orchestrator wraps it with id / segment_id / analyzer (see the orchestrator's "Findings-item shape" section) — emit only the fields below.

{
  "ambition_finding": "unambitious" | "appropriately_scoped" | "ambitious",
  "evidence": {
    "wall_clock_s": 0,
    "next_user_new_within_s": 0,
    "next_user_new_goal_overlap": "high" | "medium" | "low" | "none"
  },
  "deterministic_trigger_candidate": true | false,
  "trigger_proposal": "<if true: which external event (alert, schedule, PR open, etc.) should have fired this prompt instead of the user typing it>",
  "prompting_recommendation": "<what a single more-ambitious prompt would have looked like, or null>"
}

Evidence cites OpenTranscripts event ids (the id strings in transcript.json / segments.json), never integer turn numbers. On a root Segment — one with no following user-source New sibling — evidence.next_user_new_within_s is null and evidence.next_user_new_goal_overlap is "none"; the analyzer notes "root — no following prompt to compare" and cannot flag unambitious (there is no split-work pattern to detect).

Sequencing checklist

• Confirm segment.trigger.kind == "New" && segment.trigger.source == "user". If not, return early — this skill only applies to user-typed New Triggers.
• Pull the Segment's wall-clock from meta. If short (< a few minutes by default) and the next user-source New Trigger fires soon after and that next prompt's Goal overlaps this one's, flag as unambitious.
• If this is a root Segment with no following user-source New sibling (next_user_new_segments is empty), there is no split-work pattern to detect: set evidence.next_user_new_within_s = null, evidence.next_user_new_goal_overlap = "none", note "root — no following prompt to compare", and judge ambition_finding on the prompt's own scope alone (it cannot be unambitious on the split-work basis). The deterministic-trigger check below still applies.
• If unambitious, draft a prompting_recommendation: what one combined prompt — a single end-to-end definition of done, per philosophy-on-prompting — would have set both Goals up front?
• Independently, ask: does this user-typed New Trigger look like an ad-hoc reaction to an external event (alert, ticket, schedule, PR opening, build break)? If yes, set deterministic_trigger_candidate = true and draft a trigger_proposal naming the system that should have fired the prompt instead. This is the north-star case per the transcript-segment reference.
• For appropriately_scoped and ambitious, set the recommendation fields to null. Producing no finding is a real outcome.

Notes

• Ambition is not "longer prompts." A 1-sentence prompt that sets up a whole well-scoped piece of work is more ambitious than a 4-paragraph prompt that hand-holds the agent through every step.
• Deterministic trigger candidacy is the higher-leverage finding when it applies. The team's stated end-state is that user-typed New Triggers move toward event-triggered invocation over time; flagging candidates is how we make that gradient visible.
• Don't blame the user for splitting work that had to be split. If the next user-source New Trigger's Goal genuinely depends on the outcome of this one, that's correct scoping, not under-ambition. Use the overlap signal to distinguish them.
• Stay in the prompting bucket. This skill does not recommend Skills or MCP servers directly; if a deterministic trigger candidate implies an MCP server is needed, hand the seed to analyze-agent-transcript-mcp-gaps via the orchestrator.