Menu
Produces RUN-DECISIONS.md by filtering decision-candidate turns into real multi-alternative decisions.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
| name | narrator-decisions |
| description | Produces RUN-DECISIONS.md by filtering decision-candidate turns into real multi-alternative decisions. |
| output | json |
| inputs | ["incremental_jsonl","trace_jsonl","diff_samples"] |
You are writing the decisions artifact for a deadreckon run.
Return exactly one JSON object:
{
"design_decisions": "Markdown for choices made where the spec was ambiguous, or None.",
"deviations": "Markdown for intentional departures from the spec and why, or None.",
"tradeoffs": "Markdown for alternatives considered and why the chosen path won, or None.",
"open_questions": "Markdown for anything the owner should confirm or revise, or None.",
"decisions": [
{
"title": "short title",
"turn": 1,
"considered": ["alternative"],
"chosen": "choice",
"why": "reason with evidence",
"files_affected": ["path"],
"citations": ["[turn 1](../traces.jsonl#turn-1)"]
}
]
}
Return the four implementation interpretation fields even when no real
multi-alternative decision appears. In that case use "decisions": []; the
binary will place the no-decisions sentence under Multi-alternative decision details.
{{ incremental_jsonl }}
{{ trace_jsonl }}
{{ diff_samples }}
{{ implementation_notes }}
implementation-notes.html as source evidence for design_decisions,
deviations, tradeoffs, and open_questions.decision_candidate: true; filter false positives.files_affected must name only documentable user-authored source, config, manifest, test, asset, or project-doc files..next/, .astro/, .output/, node_modules/, .venv/, .gradle/, CMakeFiles/, .dart_tool/, .terraform/, dist/, build/, target/, .turbo/, .cache/, source maps, .env*, traces, snapshots, and run artifacts.decisions
entry. The notes feed the four interpretation fields; the decisions array
remains evidence-filtered.Amends and extends the live run narrative one beat at a time while a deadreckon run is in flight. Output is one JSON object.
Default unattended coding skill for deadreckon V0.
Produces RUN-AS-BUILT.md for a deadreckon run using stoa subsystem-AS-BUILT conventions.
Produces overview sections for deadreckon RUN-NARRATIVE.md in the stoa implementation-doc shape.
Produces phase-by-phase prose for deadreckon RUN-NARRATIVE.md from turn records and diff samples.
Produces RUN-NARRATIVE.md, RUN-AS-BUILT.md, RUN-DECISIONS.md, and optional AS-BUILT-DELTA.md from a deadreckon run trace and diff. Output is one JSON object.