From claude-commands
Builds reproduction bundles for agent failures, transcript comparisons, and retrospective analysis. Collects Claude/Codex sessions, repo evidence, hashes, and produces replay guides with LFS/encryption.
How this skill is triggered — by the user, by Claude, or both
Slash command
/claude-commands:repro-developerThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use this skill when a user wants a reproducible artifact bundle for an agent failure or cross-agent comparison, especially when raw Claude/Codex conversations, subagent transcripts, repo commits, and local state are needed.
Use this skill when a user wants a reproducible artifact bundle for an agent failure or cross-agent comparison, especially when raw Claude/Codex conversations, subagent transcripts, repo commits, and local state are needed.
Identify the incident and claim.
Collect artifacts with scripts/collect_repro.py.
Write a replay guide.
Package, do not commit large raw transcripts to an ordinary repo by default.
/tmp/repro-developer-*.Check for sensitive material before sharing externally.
SENSITIVE_REVIEW.md with findings.gitleaks detect --no-git on the sanitized bundle as the required pass/fail gate.For git-native handoff, produce two LFS artifacts.
.gpg file, suitable for Git LFS transport only; passphrase stays out of git and is shared out of band.For the July 6-7, 2026 Claude/fable vs Codex plan-miss incident, the committed artifact is:
artifacts/repro-developer/claude-fable-adversarial-review-codex-plan-miss/
It contains a sanitized LFS archive and an encrypted exact raw LFS archive. Use the artifact README first for replay.
Create a bundle:
python3 ~/.claude/skills/repro-developer/scripts/collect_repro.py \
--repo /home/$USER/projects/dark-factory \
--incident claude-adversarial-review-missed-plan-flaws \
--out /tmp \
--claude-session-dir /home/$USER/.claude/projects/-home-$USER-projects-dark-factory/61098aae-66af-474b-a488-1a47f9e8b66d \
--codex-session /path/to/codex-parent.jsonl \
--codex-session /path/to/codex-subagent.jsonl \
--repo-file docs/factory-goal-gap-review-2026-07-06.md \
--repo-file docs/adversarial-review-miss-retrospective-2026-07-06.md
Create git/LFS-ready artifacts:
python3 ~/.claude/skills/repro-developer/scripts/collect_repro.py \
--repo /home/$USER/projects/dark-factory \
--incident claude-adversarial-review-missed-plan-flaws \
--out /tmp \
--sanitize \
--require-gitleaks-clean \
--encrypt-raw \
--publish-dir artifacts/repro-developer/claude-adversarial-review-missed-plan-flaws \
...
git lfs track "artifacts/repro-developer/**"
git add .gitattributes artifacts/repro-developer/claude-adversarial-review-missed-plan-flaws
Exact command used for the committed incident:
python3 .claude/skills/repro-developer/scripts/collect_repro.py \
--repo /home/$USER/projects/dark-factory \
--incident claude-fable-adversarial-review-codex-plan-miss \
--out /tmp \
--claim "Claude fable ran a 53-agent adversarial review that confirmed current-state findings but missed plan-level hazards; a later Codex review with three subagents caught ordering, watchdog-metric, canary-overclaim, and self-reference issues already latent in the prior context." \
--claude-session /home/$USER/.claude/projects/-home-$USER-projects-dark-factory/61098aae-66af-474b-a488-1a47f9e8b66d.jsonl \
--claude-session-dir /home/$USER/.claude/projects/-home-$USER-projects-dark-factory/61098aae-66af-474b-a488-1a47f9e8b66d \
--codex-session /home/$USER/.codex/sessions/2026/07/06/rollout-2026-07-06T20-38-21-019f3aa7-c615-74a0-8d41-d643dd48a804.jsonl \
--codex-session /home/$USER/.codex/sessions/2026/07/06/rollout-2026-07-06T20-58-08-019f3ab9-e056-7540-ba77-72d4712ad4ba.jsonl \
--codex-session /home/$USER/.codex/sessions/2026/07/06/rollout-2026-07-06T20-58-14-019f3ab9-f9b6-7852-99e0-d2eba90a498d.jsonl \
--codex-session /home/$USER/.codex/sessions/2026/07/06/rollout-2026-07-06T20-58-18-019f3aba-09dd-72b2-b064-c55a372397dc.jsonl \
--repo-file docs/factory-goal-gap-review-2026-07-06.md \
--repo-file docs/adversarial-review-miss-retrospective-2026-07-06.md \
--repo-file docs/setup-agent-hooks-review-2026-07-06.md \
--repo-file roadmap/nextsteps-2026-07-06-gap-review.md \
--bead $USER-niq \
--bead $USER-ron \
--bead $USER-qdw \
--bead $USER-1m4 \
--bead $USER-gib \
--sanitize \
--require-gitleaks-clean \
--encrypt-raw \
--publish-dir artifacts/repro-developer/claude-fable-adversarial-review-codex-plan-miss
The script writes:
manifest.json with source paths, sizes, hashes, git metadata, and generated files.REPLAY.md with a developer-oriented reading order.raw/claude/ and raw/codex/ with transcripts and subagent state.repo/ with selected source documents and command outputs.checksums.sha256.<incident>.tar.zst when zstd exists, otherwise <incident>.tar.gz.--sanitize, a sibling sanitized bundle plus <incident>-sanitized.tar.*.--encrypt-raw, an encrypted raw archive and a local passphrase file that must not be committed.A bundle is complete only if it contains:
If any of those are unavailable, mark the bundle PARTIAL in REPLAY.md and list the missing evidence.
npx claudepluginhub jleechanorg/claude-commands --plugin claude-commandsRetrospects on tasks against local Codex/Claude Code transcripts to identify repeated agent failure modes and propose durable fixes in AGENTS.md or skills.
Builds an anonymised diagnostic bundle from a failed pipeline run for error reporting. Shows contents and instructs user to attach to a GitHub issue.
Structured retrospective analysis on shipped commits, releases, deployments, PRs, or sessions. Produces a 10-section report with per-fix validation, evidence sourcing, and failure-mode pattern library.