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 Sagargupta16/claude-skills --skill context-managementgit clone --depth 1 https://github.com/Sagargupta16/claude-skillsWrote 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/sagargupta16/claude-skills/context-management)<a href="https://agentmods.dev/skills/sagargupta16/claude-skills/context-management"><img src="https://agentmods.dev/badge/skills/sagargupta16/claude-skills/context-management/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/sagargupta16/claude-skills/context-management"><img src="https://agentmods.dev/badge/skills/sagargupta16/claude-skills/context-management.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 | $0.00056 | $0.01087 |
| Opus 5 | $0.00028 | $0.00544 |
| Sonnet 5 | $0.00011 | $0.00217 |
| Haiku 4.5 | $0.00006 | $0.00109 |
Grade A, and why
context-management 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 11d 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 — 116 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Context Management
Quick Reference
| Signal | Action |
|---|---|
| At ~50% context usage | Manual /compact with focus instructions |
| Approach fails, output quality drops | /rewind to before the failed attempt |
| New unrelated task mid-session | /clear or start new session |
| Need to explore/research | Spawn sub-agent (keeps main context clean) |
| Over 300K tokens used | Strongly consider new session |
| Auto-compact triggered | Too late -- quality already degraded |
Context Rot
Context rot is the gradual degradation of output quality as the context window fills up. It typically starts around 300-400K tokens (on 1M token models), but varies by task complexity.
Symptoms
- Claude repeats previous mistakes or ignores corrections
- Output becomes more generic, less project-specific
- Instructions from CLAUDE.md start being ignored
- Code patterns drift from established conventions
Prevention
- Manual compact at 50% -- don't wait for auto-compaction
- Use sub-agents for exploration tasks (research stays in child context, only results bubble up)
- New session for new tasks -- don't carry unrelated context
- Focus compaction:
/compact "keep only the auth implementation details and test results"
Branching Strategy
Every turn in a Claude Code session is a branching point. Choose deliberately:
Current turn outcome:
├── Good result → Continue (carry all context forward)
├── Mediocre result → Two options:
│ ├── /rewind (drop the attempt, retry clean)
│ └── Correct in context (adds noise, sometimes fine for small fixes)
├── Bad result / wrong direction → /rewind (always)
├── Task complete, new task → /clear or new session
└── Need deep research → Spawn sub-agent
/rewind vs Correct In Context
| Scenario | Use /rewind | Correct in context |
|---|---|---|
| Wrong approach entirely | Yes | No |
| Small typo or missed detail | No | Yes |
| Generated wrong file | Yes | No |
| Right idea, minor adjustment | No | Yes |
| Multiple failed attempts | Yes (go back further) | No |
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.
- 11d ago First seen · 116 lines · 56 tokens per session scan A f34edde85117
context-management is a skill published in the GitHub repository Sagargupta16/claude-skills (5 stars, last pushed 4d ago), licensed MIT. It adds 56 tokens to every session and 1,087 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 2026-08-31.
Other skills, from other repositories
doctor
Use to health-check the user's haku memory directory — validate file layout, catch problems before they cost data. Triggers: "run haku doctor", "check my reflection files", "is my data okay", "back up my reflections", "I'm moving to a new machine". Validates /haku-work-reflections/ (or $HAKUWORKREFLECTIONSHOME)…
capture
Use to log something from the current conversation into the user's haku memory — a decision just made, a win that surfaced, a stakeholder observation, a self-insight. Triggers: "log this", "capture this", "save that decision", "remember this", or a substantive conversation just concluded. Distinct from decision-log /…
kanban-dev-lane
Use when a Hermes Kanban worker wants to run an autonomous implementation delegation lane with a 3-tier fallback chain across claudy -> codex (--yolo) -> agyd (agy --dangerously-skip-permissions) -> Hermes direct on quota exhaustion.
release
This skill should be used when the user wants to create a new release — bump version, tag, push, create GitHub release, and optionally publish to npm. Use when user says "release", "bump version", "publish", "cut a release", or "release candidate".
council
4-voice parallel deliberation for architecture, tech selection, or design decisions with no clear answer. Each voice gets independent context to prevent anchoring bias. Use when the user says "council", "deliberate", "second opinion", or faces a trade-off between multiple valid approaches.
orbit
Autonomous pipeline — spec requirements, execute implementation, audit quality, and ship. Runs the full epic-harness orbit cycle in one go. Use when the user says "orbit", "full pipeline", "spec to ship", or describes a feature that needs end-to-end handling.