# Show Me Your Work > Produce a compact, auditable decision trail that connects requirements, evidence, choices, edits, and verification without exposing hidden reasoning. - Skill: `njs14/show-me-your-work` (Agent Skill) - Install (CLI): `npx skillmds@latest add njs14/show-me-your-work` - Raw SKILL.md: https://api.skillmd.com/api/skills/njs14/show-me-your-work/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: njs14 (https://skillmd.com/u/njs14) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/njs14/show-me-your-work --- # Show the decision trail Treat the request text that activated this skill as the work whose visible evidence should be audited. Use this method when the user requests a decision trail or the selected workflow requires one. For "show me what you did" without that context, summarize existing evidence and gaps without creating a new log. For a visual explanation, use [`show-me`](../show-me/SKILL.md). When a visual artifact accompanies a requested trail, reference that artifact's existing evidence rather than repeating its investigation. At the start, choose a stable slug. Do not create or edit the trail by hand. Append every material decision or checkpoint through the deterministic projectctl lever: ```text .pkstack/bin/projectctl evidence append \ --requirement \ --evidence \ --decision \ --verification \ --verdict VERIFIED|NOT VERIFIED|INCONCLUSIVE \ --output json ``` The default target is `.pkstack/state/evidence//decision-log.jsonl`, beneath the setup-managed ignore rule and therefore uncommitted by default. Use a committed trail only when the user explicitly requests that durable review surface, and require both `--committed` and the exact `--target Wiki/evidence//decision-log.jsonl` on every append and audit command. Neither flag alone is sufficient. An optional repository artifact requires `--artifact ` and may bind `--artifact-sha256 `; use hashes when later drift would change the claim. Each canonical line contains `schema_version`, `sequence`, `timestamp`, `requirement`, `evidence`, `decision`, `artifact`, `verification`, and `verdict`. Projectctl validates the prior file's structure, chooses the next contiguous sequence and strictly increasing timestamp, checks exact keys and bounds, validates the new event's repository-relative artifact and digest, screens known credential shapes, preserves the prior byte prefix, and replaces the file atomically while holding its lock. This is an append-only workflow convention with deterministic validation, not a signed or tamper-proof log. An old artifact may change or disappear. Append a correction without rewriting the old event; `evidence audit` still reports the historical reference's present drift. A successful append proves the write succeeded, not that every historical artifact is still current. Record decisions as they happen. Include rejected alternatives only when consequential. A retrospective reconstruction must be labeled as such and cannot claim contemporaneous coverage. Do not expose private chain-of-thought, unrelated conversation history, credentials, personal data, or unbounded raw logs. The artifact records bounded evidence and decisions, not hidden reasoning. Before handoff, run: ```text .pkstack/bin/projectctl evidence audit --output json ``` For a committed trail, repeat the exact `--committed --target Wiki/evidence//decision-log.jsonl` binding. Ask a read-only native Kiro sub-agent for an independent evidence audit when the trail is large or high risk; give it the bounded trail and named artifacts, not private transcripts. The current session reconciles all findings. Finish with a coverage summary that compares acceptance requirements with entries: proven requirements, unsupported claims, missing intervals, invalid references, verdict counts, and the exact next check.