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.
npx agentmods add skills/stablebaseline/mcp/sb-improvenpx skills add stablebaseline/mcp --skill sb-improvegit clone --depth 1 https://github.com/stablebaseline/mcpWrote 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/skills/stablebaseline/mcp/sb-improve)<a href="https://agentmods.dev/skills/stablebaseline/mcp/sb-improve"><img src="https://agentmods.dev/badge/skills/stablebaseline/mcp/sb-improve.svg" alt="Measured on agentmods" 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 | $0.00104 | $0.01831 |
| Opus 5 | $0.00052 | $0.00915 |
| Sonnet 5 | $0.00021 | $0.00366 |
| Haiku 4.5 | $0.00010 | $0.00183 |
Grade A, and why
sb-improve 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 3d 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 — 166 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Stable Baseline: improvements and risks
What this skill does
Captures and maintains the improvements backlog: improvements, risks, bugs and follow-ups, with categories, comments, activity and evidence linked back to source documents. Every change runs under the signed in user's own Stable Baseline identity.
IDs are never hardcoded. Resolve projectId, improvementId and category_id at
runtime from listing calls made in this conversation.
Before you log anything
- Resolve scope.
listOrganisations,listWorkspaces,listProjects. Ask if several projects are accessible. - Search for duplicates first.
searchImprovementsrequiresquery. Run it before creating. A backlog full of near-identical entries is worse than an empty one. If a close match exists, show it and ask whether to add evidence to that or create a new one. - Check the categories.
listImprovementCategoriesso you use an existing category rather than inventing one.
Creating
createImprovement requires projectId and title. It also takes type, priority,
status, source, urgency, category_id, problem_statement, desired_outcome,
acceptance_criteria and checklist.
- Write a real
problem_statementanddesired_outcomewhen the user has given you enough. A title alone is rarely actionable later. - Set
typeaccurately: an improvement, a risk and a bug are not the same thing and get triaged differently. - Set
sourceso the origin is traceable, for example a ticket, a meeting, a document. - Never invent a severity or priority. If the user has not said, either ask or leave it at the default and say what you did.
acceptance_criteria and checklist replace the whole array on update, and array order
is display order. Fetch with getImprovement, modify, send the whole list back, and echo
each existing row id you keep so per-row attribution survives.
Evidence
addImprovementEvidence requires improvementId and summary. It also takes
evidenceType, refId, refUrl, rawContent and position.
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.
- 3d ago First seen · 166 lines · 104 tokens per session scan A 68af69cc6330
sb-improve is a skill published in the GitHub repository stablebaseline/mcp (0 stars, last pushed 8d ago), licensed MIT. It adds 104 tokens to every session and 1,831 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.
Other skills, from other repositories
wayfinder
Plan a huge chunk of work (more than one agent session can hold) as a shared map of decision tickets on your issue tracker, and resolve them one at a time until the way to the destination is clear.
sap-transport-overview
System-wide inventory of open transport requests — every visible modifiable request, who owns it, how big it is, and supported risk signals such as empty/local requests, explicit locks, and confirmed manifest overlaps. Headers-only and cheap; NO source diffs. Use when asked "what transports are open in the system"…
timeline-creator
Create HTML timelines and project roadmaps with Gantt charts, milestones, phase groupings, and progress indicators. Use when users request timelines, roadmaps, Gantt charts, project schedules, or milestone visualizations.
pm-all
Skill "pm-all" from wei63w/pm-manager, covering user input, outline and shared workflow (all /pm- commands).
bootstrap-prd
Set up PRD-driven development infrastructure for a new project, including directory structure, templates, and roadmap.
finn-spec
Interview the user about a raw idea until confident, then file a build-ready issue in Linear. Use when asked to run Finn-loop's spec interview, draft a queue-ready issue, or plan a feature. Interactive — requires the user present; never run unattended.