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 lukasrepublic/agentic-foundry --skill learn-distillgit clone --depth 1 https://github.com/lukasrepublic/agentic-foundryWrote 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/lukasrepublic/agentic-foundry/learn-distill)<a href="https://agentmods.dev/skills/lukasrepublic/agentic-foundry/learn-distill"><img src="https://agentmods.dev/badge/skills/lukasrepublic/agentic-foundry/learn-distill.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 5.1 | $0.00077 | $0.00709 |
| Opus 5 | $0.00039 | $0.00354 |
| Sonnet 5 | $0.00015 | $0.00142 |
| Haiku 4.5 | $0.00008 | $0.00071 |
Grade A, and why
learn-distill 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 4d 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 — 43 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/foundry:learn-distill
The self-improvement consumer — a native-scheduling WRAP for the cadence. Two halves:
- CUSTOM (no native equivalent): the deterministic clustering + report logic, implemented in
scripts/foundry-distill.py— read the dated partitions under.foundry/session-learnings/<YYYY-MM-DD>/*.jsonl(the reconciliation.harvest-log.jsonl+ dotfiles excluded), token-overlap cluster (≥3 shared normalized tokens drawn from every string scalar field +tags; no embeddings; fully-pinned canonicalization + ordering → byte-reproducible), threshold-promote (≥3-record components) to HBK/memory/skill/unclassified candidates over the realtype/kindvocabulary, write.foundry/learnings/DISTILL_REPORT-<UTC-date>.md(an untracked runtime artifact — NOT the governedspecs/tree), then prune buffer partitions whose directory-NAME date is >90 days old (never by mtime). Run it directly:scripts/foundry-distill.py [--date YYYY-MM-DD] [--dry-run]; self-testscripts/foundry-distill.py --selftest. (The doctor drop-in check this line originally named was retired with the drop-in registry in the v0.25.0 realignment — an open loop is now visible only by running the distiller itself.) - WRAP over native: the CADENCE. Instead of a bespoke scheduler, the distill tick
is scheduled via native
ScheduleWakeup(session-local) orCronCreate(persistent) — e.g. a daily tick that runs the consumer. Producers feed the buffer via/foundry:learn-capture(the in-session path) + the worker-sidecar harvest.
Procedure
- Scheduled tick (native
CronCreatedaily, orScheduleWakeup): invoke the consumer —scripts/foundry-distill.py. - Consume the buffer (Read → Cluster → Threshold → Write → Retention, in that strict order — retention prunes only partitions whose NAME-date is >90 days old, never a within-window partition).
- Emit
.foundry/learnings/DISTILL_REPORT-YYYY-MM-DD.mdwith HBK/memory/skill/unclassified candidates (operator-reviewed; no auto-dispatch).
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.
- 4d ago First seen · 43 lines · 77 tokens per session scan A 36159ecda5b6
learn-distill is a skill published in the GitHub repository lukasrepublic/agentic-foundry (1 stars, last pushed 2d ago), licensed MIT. It adds 77 tokens to every session and 709 once invoked, about $0.0004 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 2026-09-03.
Other skills, from other repositories
capture-milestone-principle-updates
Distill reusable answering principles from a just-finished milestone's recorded decisions into the project-wide principle store.
dual-brain
Two complementary reusable sub-agents collaborate to remember, grill, verify, document, and maintain weighted project memory. A Right Brain (context / pattern / grill) interrogates assumptions against Hot/Warm memory; a Left Brain (logic / verification / code) cross-checks memory, code, and docs; the orchestrator…
yolo-intake
Use when a feature should draw on the project's reference material for context. yolo-intake does NOT fetch, import, or copy anything — it points YOLO at a selected folder (default docs/) and READS from it to enrich a feature's context. Getting material INTO that folder (a Figma, a spec, a URL, a schema) is the docs…
g-resume
Re-hydrate a fresh session with the right slice of the durable record. The read-side counterpart to /g-retro — selectively retrieves the relevant retro, ADRs, journal, and handoff keyed by the current branch/milestone/first-task, and assembles a focused re-entry briefing. Loads distilled context into a clean window…
grimoire
Use when the user says 'update context', 'update claude', 'save library', or after significant project changes.
continuous-learning-v2
Instinct-based learning system that observes sessions via hooks, creates atomic instincts with confidence scoring, and evolves them into skills/commands/agents.