codeops_retro

A terminal report that reviews telemetry from CodeOps quality checks over a chosen time period. Telemetry is recorded information about what those checks found and how people responded.

In plain words
What is it for?
Use it for periodic reviews after releases or several work phases to decide whether to adjust the agents or the repository's quality settings.
Why use it?
It helps show when quality agents are producing unhelpful results or when expected checks are missing, without changing the project automatically.

Command

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add commands/blendsdk/claude-codeops/codeops_retro
Clone the repo
git clone --depth 1 https://github.com/blendsdk/claude-codeops
Per session 96 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,157 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00096 $0.01157
Opus 5 $0.00048 $0.00579
Sonnet 5 $0.00019 $0.00231
Haiku 4.5 $0.00010 $0.00116

Measured 2d ago against content hash 8b04ac9e64d7, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

codeops_retro scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

commands/codeops_retro.md · 81 lines

How it starts

The opening of the file, as written. The whole thing — 81 lines — stays where its author put it; the contents beside it link to each section on GitHub.

codeops_retro — the quality loop's own review

Judge whether the quality agents are earning their keep, from the telemetry alone. Cadence: monthly OR roughly every 10 executed phases, whichever comes first — plus one focused retro shortly after each release. This command is terminal-only (no file output) and recommend-only: it proposes changes, and the user applies them.

Step 1 — Gather

Run the utility at "${CLAUDE_PLUGIN_ROOT}/scripts/codeops-events.sh" (the sole reader of the events file — never open events.jsonl yourself):

  1. stats --since <Nd> (default 30d, or the user's window)
  2. stats --since <Nd> --by agent
  3. stats --since <Nd> --by lens
  4. stats --since <Nd> --by delivery
  5. stats --since <Nd> --by drift
  6. stats --since <Nd> --by design
  7. gaps --since <Nd>

Relay each table verbatim, then analyze.

Step 2 — Apply the thresholds (this table is the owning copy)

Signal Trigger Reading
Per-agent acceptance rate < 40% with n ≥ 10 rulings The agent's findings are mostly rejected — its prompt, checklists, or severity calibration need tuning
Emission gap rate > 20% Reviews complete without recorded rulings — the dispatch wiring or ruling discipline is leaking
Blocker category same category ≥ 3× in the window A systemic weakness in packets or plans, not bad luck
Lens acceptance a lens with 0 accepted findings over n ≥ 10 phases The lens adds noise, not value — candidate to drop from the profile
First-pass verify rate < 60% with n ≥ 10 tasks Tasks arrive at verification unfinished — task granularity or the spec tests, not the executor
Spec tests failing post-impl > 15% of authored, n ≥ 20 Tests confirmed red are passing for the wrong reason — a spec-author calibration problem
Re-review rate > 40% with n ≥ 10 phases Phases routinely need a second pass — the dispatch packet is not carrying enough
Runtime ambiguity owner one stage ≥ 50% with n ≥ 10 That stage is closing over decisions it should be settling
Resume marks corrected > 10% with n ≥ 10 resumes The two-stage [~]/[x] mark is not surviving interruption
Delegated escalation share > 50% with n ≥ 10 decisions Delegated design is moving questions rather than answering them; split by escalation reason before judging
gate_summary trends display-only Context only — no gate-friction threshold yet (deliberately deferred until about a quarter of gate data exists)

Read the full file on GitHub · 81 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 81 lines · 96 tokens per session scan A 8b04ac9e64d7

Subscribe to this mod's changes

codeops_retro is a command published in the GitHub repository blendsdk/claude-codeops (4 stars, last pushed 1mo ago), licensed MIT. It adds 96 tokens to every session and 1,157 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.