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 justinjdev/fellowship --skill lembasgit clone --depth 1 https://github.com/justinjdev/fellowshipWrote 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/justinjdev/fellowship/lembas)<a href="https://agentmods.dev/skills/justinjdev/fellowship/lembas"><img src="https://agentmods.dev/badge/skills/justinjdev/fellowship/lembas/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/justinjdev/fellowship/lembas"><img src="https://agentmods.dev/badge/skills/justinjdev/fellowship/lembas.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 | $0.00051 | $0.01171 |
| Opus 5.5 | $0.00020 | $0.00468 |
| Sonnet 5.5 | $0.00010 | $0.00234 |
| Haiku 4.5 | $0.00005 | $0.00117 |
Grade A, and why
lembas 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 — 120 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Lembas — Intentional Context Compression
Overview
Compresses the current conversation into a structured summary to keep the context window in the "smart zone." This is the intentional compaction pattern from context engineering — proactively trimming context between phases rather than waiting for overflow.
Reasoning quality degrades as a context window fills — well before it overflows. This skill is the mechanism for staying in the sweet spot by compacting early and often.
When to Use
- Between phases of the quest workflow (invoked automatically by
/quest) - After any stretch of verbose output you would not re-read (build logs, long file reads, exploratory searches, failed-attempt noise) — if in doubt, compact
- Before switching focus within a session
- Standalone via
/lembas
Process
Step 1: Identify Current Phase
Determine what just completed:
- Research: Understanding the system, identifying files
- Plan: Outlining steps, getting approval
- Implement: Writing code, running tests
- Review: Adversarial review, conventions, verification, PR
- Ad hoc: No formal phase — general work
Step 2: Extract Essentials
Review the conversation and extract only what matters for the next phase. Be aggressive about discarding noise:
Keep:
- Decisions made and their rationale
- Files identified with specific line ranges
- Constraints discovered
- Open questions that still need answers
- Test results (pass/fail, not full output)
Discard:
- Raw grep/search output (keep only the conclusions)
- Full file contents (keep only relevant line ranges)
- Verbose build/test output (keep only the verdict)
- Exploratory dead ends (keep only what was learned)
- Repeated information
Step 3: Produce Compacted Context Block
Output in this exact format:
## Compacted Context
### Phase Completed: [Research | Plan | Implement | Review | Ad hoc]
### Task
[one-line description, carried forward from Session Context]
### Package(s)
[package name(s) and path(s), carried forward from Session Context]
### Key Findings
- [decisions made this phase]
- [constraints discovered]
- [patterns identified]
### Files
- [file:lines] — [what's relevant and why]
- [file:lines] — [what's relevant and why]
### Current State
- [what's been done so far]
- [what's working / what's broken]
### Next Phase
- [what needs to happen next]
- [open questions to resolve]
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 Changed · +4 lines f86190fd8570
- 1mo ago Changed · +3 lines · +6 tokens per session 6e82f38ccb27
- 1mo ago First seen · 113 lines · 45 tokens per session scan A ed287fb8d463
lembas is a skill published in the GitHub repository justinjdev/fellowship (5 stars, last pushed 16d ago), licensed Apache-2.0. It adds 51 tokens to every session and 1,171 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
continuous-learning
ProductionOS instinct-based learning system. Observes sessions via hooks, extracts patterns, creates atomic instincts with confidence scoring, and auto-promotes high-confidence patterns across projects.
cross-task-learner
Enable agent loops to learn from similar past tasks and share patterns across loops.
ralph-memory
Manage Al semantic memory entries — list, query, and clear lessons learned across loop iterations.
session-explore
Investigate past AI session activity with cited catalog search, timelines, tool analytics, and bounded comparisons across providers.
reflection-injection
Inject relevant past reflections into agent context at iteration start so agents learn from prior mistakes without repeating them.
session-harvest
Extract cited decisions, requirements, risks, entities, and relationships from session history and review exact candidate versions before memory promotion.