agentsope/SkillAlchemy / skills/agentsop-session-state-hygiene
agentsop-session-state-hygiene
VerifiedDecision protocol for managing the context/session state of an AI coding tool: when to /clear, when to keep context, and how to detect "context bleed" — the failure mode where stale conversation history biases the model against the current task. Surfaces a discipline that Aider (/clear), Claude Code (/clear), CrewAI (memory=False, re-instantiate), and LangGraph (new thread_id, subgraph isolation) all encode separately but none name as a skill.
| File | Kind | Size |
|---|---|---|
| README.md | markdown | 3.2 KB |
| SKILL.md | skill md | 25.3 KB |
| intermediate/operation_candidates.json | config | 10.7 KB |
| references/R1-source-evidence.md | markdown | 5.6 KB |
| references/R2-tool-commands.md | markdown | 6.3 KB |
README.md
View on GitHub# session-state-hygiene
A Phase-D, common-tier, daily-frequency coder-agent skill: **when to `/clear`,
when to keep context, and how to detect "context bleed."**
> One line: context is signal; stale context is noise; clearing restores signal.
## What this is
A multi-turn AI coding session is a sliding window of evidence. Early on it's
pure signal; the longer it runs, the more dead ends, abandoned plans, and
superseded files accumulate — until yesterday's good context becomes today's bad
bias ("context bleed"). This skill is the discipline of noticing that moment and
acting with the smallest correct cut.
Aider (`/clear`), Claude Code (`/clear`), CrewAI (`memory=False`, re-instantiate)
and LangGraph (new `thread_id`, subgraph isolation) each encode this discipline
separately. None names it as a skill. This one does, and maps the single move
onto every tool's command.
## When to activate
- **Topic shift** inside a session (finished A, starting unrelated B).
- **Length warning** — window N% full, or Aider `/tokens` past ~25k (the
observed point where models start ignoring their system prompt).
- **Weird behavior** — model repeats a corrected mistake, cites a removed
file/decision, or obeys an older instruction over the newest one.
Do **not** activate for a single call, a one-shot RAG query, or a clean
on-topic window — hygiene on clean context is superstition.
## The SOP, in four steps
1. **Recognize** — name the symptom in one line before touching anything.
2. **Save what's worth** — externalize keepers (commit code, write the plan to a
note) so they survive the clear.
3. **Clear** — the smallest correct cut: drop one item / trim history / clear
history / fresh session.
4. **Restart focused** — open the clean window with a one-paragraph summary, not
the old transcript.
## Files
- `SKILL.md` — the skill. 7 sections: 何时激活 / 核心心智模型 / SOP / 操作模型
(OP-001..OP-009) / 困境决策案例 (3 cases) / 反模式与边界 / 跨框架对照.
- `references/R1-source-evidence.md` — primary sources, grouped by claim, with
full URLs for the inline `[tool/topic]` citation tags.
- `references/R2-tool-commands.md` — copy-pastable clear/reset commands for
Aider, Claude Code, CrewAI, LangGraph, web UIs, and raw SDK loops.
- `intermediate/operation_candidates.json` — machine-readable operations and
anti-patterns the SKILL.md was distilled from.
## Cross-framework cheat sheet
| Need | Aider | Claude Code | CrewAI | LangGraph |
|---|---|---|---|---|
| Inspect size | `/tokens` | `/context` | trace | `len(state["messages"])` |
| Clear, keep files | `/clear` | `/clear` | re-instantiate `Crew()` | new `thread_id` |
| Hard reset | `/reset` | exit CLI | new process + wipe store | new `thread_id` |
| Partial clear | `/drop <files>` | n/a | trim `context=[...]` | `update_state({"messages":...})` |
| Isolate sub-task | 2nd window | sub-process | separate `Crew`, `memory=False` | subgraph |
Full commands in `references/R2-tool-commands.md`.
## Provenance
Distilled from three local sibling SOPs — `aider-sop`, `crewai-sop`,
`langgraph-sop` — plus vendor docs. Every framework-specific claim is cited
inline and resolved in `references/R1`.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |