Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/iliaal/whetstonenpx agentmods add commands/iliaal/whetstone/diagnose-negativesWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/commands/iliaal/whetstone/diagnose-negatives)<a href="https://agentmods.dev/commands/iliaal/whetstone/diagnose-negatives"><img src="https://agentmods.dev/badge/commands/iliaal/whetstone/diagnose-negatives/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/commands/iliaal/whetstone/diagnose-negatives"><img src="https://agentmods.dev/badge/commands/iliaal/whetstone/diagnose-negatives.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00022 | $0.01772 |
| Opus 5 | $0.00011 | $0.00886 |
| Sonnet 5 | $0.00004 | $0.00354 |
| Haiku 4.5 | $0.00002 | $0.00177 |
Grade A, and why
diagnose-negatives 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 10d 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.
How it starts
The opening of the file, as written. The whole thing — 112 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Diagnose and fix skill failures
Read real negative-signal sessions for a skill, identify recurring failure patterns, and propose targeted skill text edits. This is the practical alternative to automated evolution -- it reads what actually went wrong and fixes the specific gaps.
Arguments
SKILL_NAME=$1 (required)
Parse from: $ARGUMENTS
If no skill name provided, run analyze-misfires first to show which skills have the most issues, then ask which one to diagnose.
Pipeline
Step 1: Run diagnosis
python3 distillery/scripts/distiller.py harvest-sessions
Run the diagnosis judge as an in-session sub-agent (no billed claude -p):
python3 distillery/scripts/distiller.py diagnose-negatives <skill> --max-examples 10 --emit-prompt→{count, prompt, meta}. Ifcountis 0, stop — no negatives to diagnose.- Spawn ONE sub-agent (Agent tool,
general-purpose) whose entire instruction is theprompt; it returns the JSON diagnosis. Write that response to a temp file. python3 distillery/scripts/distiller.py diagnose-negatives <skill> --max-examples 10 --format-result --response @<file>→ validates against the rubric and prints the report.
The judge classifies each finding under exactly one of seven smallest-failing-decision categories:
| Category | Failure mode | Default edit target |
|---|---|---|
wrong_trigger |
Skill fired when not relevant, or didn't fire when it should | hooks/skill-patterns.sh regex; SKILL.md description |
missing_source |
Skill ran but didn't load a reference it needed | Add or expand references/; update SKILL.md routing table |
skipped_reference |
Reference exists, agent had access, didn't read it | Tighten SKILL.md routing; promote content into SKILL.md |
weak_output |
Output format loose or missing structure | Add output template / table format to SKILL.md |
missing_validation |
Skill claimed completion without running a check | Add gate to SKILL.md; add validator check if catchable |
unsafe_path |
Destructive action without confirmation | Add confirmation requirement; ban destructive verbs |
other |
None of the above; only valid with deferred_reason |
Manual decision |
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.
- 10d ago First seen · 112 lines · 22 tokens per session scan A 89bee2cfee29
diagnose-negatives is a command published in the GitHub repository iliaal/whetstone (33 stars, last pushed 2d ago), licensed MIT. It adds 22 tokens to every session and 1,772 once invoked, about $0.0001 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 2026-08-30.
Other commands, from other repositories
pr-review
Comprehensive PR review across six dimensions — cost impact, environment drift, ownership gaps, SOC 2 compliance, deprecated API / version hygiene, and rollback feasibility. Each mode inspects the diff and current file state, reports findings with severity, and recommends concrete fixes. Use when preparing a PR for…
triage
Triages a PR comment — from a bot (Copilot, CI) or a human reviewer. Fetches the comment and diff via gh CLI, classifies it, applies the fix directly to the file if valid, posts a reply on the thread, and resolves it. Run from inside the repo.
review
Multi-agent code review with parallel validation.
sm-buff
You usually don't run buff directly — run sm sail. sm sail drives the whole PR to green and calls buff watch / triage for you, stopping only when it needs you to act (see /sm-sail). Reach for sm buff here only for surgical work: inspecting a specific failure, or resolving a single review thread when sail has parked on…
sm-scour
Run slop-mop's comprehensive pre-PR sweep for this repository.
speckit.boundedrelay.implementation-review
Run frozen Claude host and independent Codex implementation reviews.