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 skills add tzachbon/smart-ralph --skill ralph-specum-implementgit clone --depth 1 https://github.com/tzachbon/smart-ralphWrote 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/tzachbon/smart-ralph/ralph-specum-implement)<a href="https://agentmods.dev/skills/tzachbon/smart-ralph/ralph-specum-implement"><img src="https://agentmods.dev/badge/skills/tzachbon/smart-ralph/ralph-specum-implement.svg" alt="Measured on agentmods" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00053 | $0.01195 |
| Opus 5 | $0.00026 | $0.00598 |
| Sonnet 5 | $0.00011 | $0.00239 |
| Haiku 4.5 | $0.00005 | $0.00120 |
Grade A, and why
ralph-specum-implement 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 8d 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 — 56 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Ralph Specum Implement
You are a coordinator, not an executor -- delegate each task to a spec-executor sub-agent.
Contract
- Resolve the active spec by explicit path, exact name, or
.current-spec - Require
tasks.md - Recompute task counts from disk before execution
- Merge state fields only
- Reconcile prototype records before dispatch and block only dependent work
- Remove
.ralph-state.jsononly when all tasks are complete, verified, andactivePrototypesis empty
Action
- Resolve the active spec. If none exists, stop.
- Require
tasks.md. Read.progress.md, current state, and current task markers. - Parse
tasks.mdonce into ordered top-level task rows. Include only unindented checkboxes outside fenced example blocks whose next token is a concrete numeric task ID,V<number>,VE<number>, orVF; exclude nested and example checkboxes, completion criteria, and placeholder IDs. From that one list derivetotal, the completed count across all rows, andnext_indexas the zero-based position of the first incomplete row ortotalwhen none remains. Do not derivenext_indexfrom the completed count; non-prefix completion cases resume at the earliest incomplete row. - Resolve the dispatch task index before merging state. For fresh execution, use
next_index. For a prototype return, require a validated non-negativereturnTaskIndexand verify that it identifies the first eligible incomplete task. Merge state once with:phase: "execution"awaitingApproval: falsetotalTasks: total- taskIndex:
next_indexfor fresh execution, or the validatedreturnTaskIndexfor a prototype return - preserve
taskIteration,maxTaskIterations,globalIteration,maxGlobalIterations,commitSpec, andrelatedSpecs
- Before dispatch, run
prototype_records.py reconcilewhenever state exists and runselect-downstreamwheneveractivePrototypesis nonempty or prototype history exists. Request--target execution,--target "task:$TASK_INDEX", and--pathfor every declared current-task path. Stop when an active blocker or stale input targets the work, or when any matchingtargetDecisionsentry is not bothproofAvailable: trueandeligible: true. Missing dependency or approved-transfer proof blocks conservatively. Report the prototype ID and resume active work through$ralph-specum-prototype --resume <id>; route terminal staleness to its earliest affected phase or task. - On a prototype return, verify that the merged
taskIndexstill equals the validatedreturnTaskIndexand identifies the first eligible incomplete task before dispatch. - Delegate each task to a
spec-executorsub-agent. Pass the task description, file targets, success criteria, and context from.progress.md. The sub-agent implements the task and outputsTASK_COMPLETE. Do NOT implement tasks yourself. Execute tasks in order until complete or blocked. [P]tasks may batch only when file sets do not overlap and verification is independent.[VERIFY]tasks stay in the same run and must produce explicit verification evidence.- Marker syntax must be explicitly present in
tasks.md. If markers are absent, treat tasks as non-batchable by default. - VE tasks are valid quality tasks when the spec includes autonomous end-to-end verification.
- Native task sync metadata should be preserved when present.
- After each task or safe batch:
- mark the checkbox
- update
.progress.md - merge the state update
- use the task
Commitline unless commits were explicitly disabled
- Before any batching, generated-task, CI, review-fix, branch-publication, or PR-lifecycle push, apply the Prototype Evidence Push Gate in
../../references/workflow.md. Normal mode may ask at that boundary for separate explicit authorization naming every outbound**/prototypes/*.mdrecord. Quick mode asks no question and skips every push. A skipped or denied push ends the dependent remote lifecycle path: do not rungh pr create,gh pr merge,gh pr checks,gh pr view,gh api,gh run,gh issue, remote review polling, issue writes, or later remote steps that depend on that push. Quick mode continues or finishes locally and reportsRemote lifecycle skipped: prototype evidence stayed local.Preserve the existing normal remote lifecycle only after the gate completes the push. Never push an isolated prototype source branch.commitSpecauthorizes local commits only. - On failure or interruption, persist the current state and stop with a resumable summary.
- On full completion, reconcile again. If
activePrototypesremains nonempty, preserve.ralph-state.jsonand stop with its IDs. Otherwise remove state and report completion.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 8d ago First seen · 56 lines · 53 tokens per session scan A 143ef23ca703
ralph-specum-implement is a skill published in the GitHub repository tzachbon/smart-ralph (535 stars, last pushed 4d ago), licensed MIT. It adds 53 tokens to every session and 1,195 once invoked, about $0.0003 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 skills, from other repositories
remember
Review reusable project knowledge and decide what belongs in project memory, notepad, or durable docs.
embedding-strategies
Select and optimize embedding models for semantic search and RAG applications. Use when choosing embedding models, implementing chunking strategies, or optimizing embedding quality for specific domains.
docx-comment-reply
Reply to comments (批注) in Word .docx/.doc files: extract comment context, draft replies, write threaded replies back, and validate OOXML.
explaining-machine-learning-models
Explain trained machine learning models through feature attribution, local explanations, and behavior summaries. Use as an explicit/manual helper once a model already exists, not for training ownership, leakage auditing, or general ML strategy selection.
aiwg-regenerate-copilot
Regenerate copilot-instructions.md for GitHub Copilot with vendor-specific content only.
steward-prep-delivery
Steward-assisted prep for filing issues and PRs — environment capture, template selection, duplicate detection, delivery-policy compliance check.