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 Obsidian-Owl/specwright --skill sw-learngit clone --depth 1 https://github.com/Obsidian-Owl/specwrightWrote 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/obsidian-owl/specwright/sw-learn)<a href="https://agentmods.dev/skills/obsidian-owl/specwright/sw-learn"><img src="https://agentmods.dev/badge/skills/obsidian-owl/specwright/sw-learn/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/skills/obsidian-owl/specwright/sw-learn"><img src="https://agentmods.dev/badge/skills/obsidian-owl/specwright/sw-learn.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 | $0.00040 | $0.01986 |
| Opus 5.5 | $0.00016 | $0.00794 |
| Sonnet 5.5 | $0.00008 | $0.00397 |
| Haiku 4.5 | $0.00004 | $0.00199 |
Grade A, and why
sw-learn 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 1mo 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 — 150 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Specwright Learn
Goal
Extract reusable knowledge from the current work unit. Build failures, gate findings, and architecture decisions contain valuable patterns. Surface them, let the user curate, and promote the best ones so future work benefits.
Inputs
{worktreeStateRoot}/session.json-- selected work for this worktree{repoStateRoot}/work/{selectedWork.id}/workflow.json-- current work unit (should be shipped){workDir}/evidence/-- gate evidence files{workDir}/implementation-rationale.md-- curated build-time reasoning when present{workDir}/review-packet.md-- reviewer-facing audit synthesis when present{workDir}/plan.md-- architecture decisions{workArtifactsRoot}/{selectedWork.id}/approvals.md-- approval lineage for the work{projectArtifactsRoot}/CONSTITUTION.md-- existing practices{projectArtifactsRoot}/learnings/-- prior work unit learnings (for retrospective)- Git log for the work unit's commits
Outputs
- Learnings presented to user in categories
- User-approved patterns promoted to one of:
{projectArtifactsRoot}/CONSTITUTION.md(new practice rule)- Auto-memory MEMORY.md (compact pattern entry, loaded every session)
{projectArtifactsRoot}/patterns.md(reusable pattern library)
{projectArtifactsRoot}/learnings/{work-id}.json-- written when any finding is promoted OR when gateCalibration data is available. When only calibration is present (no promoted findings), write with an emptyfindingsarray.
Constraints
Stage boundary (LOW freedom):
- Follow
protocols/stage-boundary.md. - You capture learnings and promote patterns. You NEVER start new work units, run builds, or create PRs.
- It is not a core workflow stage and never claims top-level work ownership.
- After learnings are captured, STOP and present the handoff:
- If more work units pending: "Run
/sw-buildto start the next unit." - If no more units: "All work units complete. Learnings captured."
- If more work units pending: "Run
Discovery (HIGH freedom):
- Scan evidence files, git log, plan.md,
implementation-rationale.md,review-packet.md, and approval lineage for patterns worth remembering. - Look for: what broke, what was hard, what worked well.
- Check as-built notes for discovered behaviors per
protocols/build-quality.md. - If
commands.test:integrationis configured in config.json: check gate-build evidence for the integration tier. If it was SKIP or absent, surface as a learning candidate ("No integration tests ran"). Skip this check when no integration tier is configured. - MUST record gateCalibration for every gate that ran, even if all PASS with 0 findings. Populate from evidence files automatically. falsePositives array only populated when user explicitly labels a finding as false positive during presentation (dismissal alone does not count). Format per
protocols/evidence.md#verdict-rendering.
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.
- 1mo ago First seen · 150 lines · 40 tokens per session scan A 49d29d640b77
sw-learn is a skill published in the GitHub repository Obsidian-Owl/specwright (10 stars, last pushed 5mo ago), licensed MIT. It adds 40 tokens to every session and 1,986 once invoked, about $0.0002 per session on Opus 5.5. 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
doc-digest
Turn a long document, PDF, transcript, spreadsheet or log into a compact working digest the conversation can continue from. Use when the user shares or pastes something long (several pages or more) and wants to discuss, analyze, compare or use it over more than one message, or asks to summarize a document so they can…
fresh-start
Continue in a new chat without losing anything — writes a compact handoff to paste into a fresh conversation. Use whenever the user wants to move to a new chat or conversation, hits usage limits, says replies got slow or Claude forgets earlier details, or the conversation has grown long.
token-check
Estimate how many tokens a pasted text, an attached file or the conversation so far takes, what it means for the context window and usage limits, and how to cut it. Use when the user asks how big something is in tokens, why they keep hitting limits, or whether a file is too large to send.
project-context
Turn background the user keeps re-explaining or re-pasting into reusable instructions, so new chats start with it for free. Use when the user repeats the same context, rules, style guide, persona or reference text across messages or chats, says they are tired of explaining the same thing, or asks how to set up a…
performance-audit
Full-stack performance health check across 12 dimensions. Rendering, bundles, assets, API/network, algorithms, memory, database, caching, Web Vitals, backend runtime, concurrency, and framework-specific pathologies. Evidence-based Impact Models with confidence tiers and a prioritized optimization roadmap. Switches…
sitrep
Situational awareness — where am I, what was I doing, what's next. Context recovery after compression, confusion, or mid-session reorientation.