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 ychampion/cskill-agents --skill progressive-compaction-promptgit clone --depth 1 https://github.com/ychampion/cskill-agentsWrote 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/ychampion/cskill-agents/progressive-compaction-prompt)<a href="https://agentmods.dev/skills/ychampion/cskill-agents/progressive-compaction-prompt"><img src="https://agentmods.dev/badge/skills/ychampion/cskill-agents/progressive-compaction-prompt/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/ychampion/cskill-agents/progressive-compaction-prompt"><img src="https://agentmods.dev/badge/skills/ychampion/cskill-agents/progressive-compaction-prompt.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 | $0.00028 | $0.00523 |
| Opus 5 | $0.00014 | $0.00262 |
| Sonnet 5 | $0.00006 | $0.00105 |
| Haiku 4.5 | $0.00003 | $0.00052 |
Grade A, and why
progressive-compaction-prompt 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 — 30 lines — stays where its author put it; the contents beside it link to each section on GitHub.
SKILL: Progressive Compaction Prompt
Domain: context-management Trigger: Apply when composing compact summaries so each round captures intent, context, per-step detail, and invocation cues before the final summary is stripped to user-friendly text. Source Pattern: Distilled from reviewed session memory, compaction, and context-budgeting implementations.
Core Method
Wrap each compaction request in a staged prompt that asks the model to summarize in a deliberate sequence instead of freeform prose. The prompt should first ban tool use, then walk through the important dimensions of the conversation such as goals, decisions, artifacts, open work, and invocation guidance. Ask the model to emit a structured draft that is easy for post-processing to clean up, then strip the drafting scaffolding before storing or showing the final summary. Add only the minimum follow-up instructions needed so an automated agent can resume from the compacted state without asking redundant questions.
Key Rules
- Use a dedicated no-tools preamble so the compaction agent never tries to call tools while summarizing.
- Pick the prompt template based on scope: one template for full-session compaction and a different one for partial compaction that preserves recent turns verbatim.
- Keep the structured draft easy to clean: analysis scaffolding may exist temporarily, but the stored or displayed summary should be post-processed into plain readable text.
- Include only the follow-up instructions that downstream automation actually needs, such as whether to suppress clarification questions or where to find preserved transcript artifacts.
- Reuse the same prompt shape across compaction flows so summaries remain consistent enough for tooling to parse and compare.
Example Application
Before resuming a long-running session after compaction, generate the summary with the staged prompt, run a cleanup pass that removes drafting tags, and attach the final summary to the resumed session along with any preserved transcript references. The next agent can then continue from a concise, automation-friendly handoff instead of reprocessing the trimmed history.
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 · 30 lines · 28 tokens per session scan A 04927af707ca
progressive-compaction-prompt is a skill published in the GitHub repository ychampion/cskill-agents (36 stars, last pushed 5mo ago), licensed MIT. It adds 28 tokens to every session and 523 once invoked, about $0.0001 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
grape
Use Grape MCP for Codex context continuity in coding repositories. Use when a task needs repeated-turn context, omitted context restore, stale-context checks, invalidation checks, or safe continuity across branch and dirty-worktree changes.
learn
Diagnose and fix agent behavioral surfaces when the user corrects a mistake — connects to Claude native memory.
ackit-context-optimization
Build budgeted context packs and trim instruction bloat for coding agents.
chain
Inject the 3 most recent session IDs for the current workspace into the prompt before Claude responds. Use when starting work and wanting recent session context.
context-efficient-coding
Use ContextFlux to minimize coding-agent context and tool calls when exploring, planning, debugging, reviewing, or modifying an unfamiliar or multi-file repository. Trigger for codebase discovery, symbol search, architecture questions, cross-file changes, error tracing, review feedback, test discovery, or any request…
k-knowledge
Instructions for recording stable project facts in AGENTS.md or a project map. AGENTS.md is a file that stores guidance for coding agents.