Agent skill
context-preservation
Generate structured context summaries for session continuity across compaction
Install this agent skill to your Project
npx add-skill https://github.com/juanandresgs/claude-ctrl/tree/main/skills/context-preservation
SKILL.md
Context Preservation Skill
Generate structured context summaries when compaction is imminent or requested, ensuring session continuity without information loss.
Output Format (STRICT - NO DEVIATIONS)
### 1. Current Objective & Status
- **Goal**: [Robust, multi-sentence description of what we are building. Include critical details, context, and the "Definition of Done". Do not summarize into a single vague line.]
- **Status**: [Completed | In Progress | Blocked]
- **Immediate Next Step**: [The very next command or code edit required. Be specific.]
### 2. Active Context
- **Active Files**:
- `[Absolute Path]`
- `[Absolute Path]`
- **Recent Changes**: [Specific descriptions of what *just* changed in the code, referencing specific functions or logic.]
- **Variables/State**: [Key variable names, data structures, or temporary states currently in focus.]
### 3. Constraints & Decisions (CRITICAL)
- **Preferences**: [User preferences stated in this session, e.g., "no external deps", "use snake_case"]
- **Discarded Approaches**: [What did we try that failed? Do not repeat these mistakes.]
- **Architectural Rules**: [Patterns we agreed on, e.g., "Service Layer pattern", "DTOs required"]
### 4. Continuity Handoff
- "When resuming, the first thing to do is..."
Content Extraction
Section 1: Current Objective & Status
- Extract goal from user's initial request, explicit goal statements, Definition of Done indicators
- Status:
Completed(all objectives achieved),In Progress(work ongoing),Blocked(explicit blocker) - Next step: last in-progress item, last "next I will..." statement, or first pending objective
Section 2: Active Context
- Files touched via Read/Write/Edit operations, prioritized by recency
- ALWAYS use absolute paths (e.g.,
/Users/turla/Code/project/src/auth.ts) - Summarize function additions, modifications, deletions with specific names and line refs
- Key variables, data structures, environment variables in focus
Section 3: Constraints & Decisions
- Preferences: "don't use...", "prefer...", "always...", "never..."
- Discarded approaches: failed attempts with reasons
- Architectural rules: pattern agreements, structure decisions, dependency choices
Section 4: Continuity Handoff
- Always begin with "When resuming, the first thing to do is..."
- Must be specific enough for a fresh Claude instance to immediately continue
Anti-Patterns (NEVER DO)
Goal Description
WRONG:
- "Building a feature"
- "Working on authentication"
RIGHT:
- "Building a user authentication system with OAuth2 PKCE flow for the mobile app. The system must support Google and Apple SSO, store refresh tokens securely in the device keychain. Definition of Done: user can complete full login flow, tokens persist across app restarts, logout clears all tokens."
Active Files
WRONG:
file.ts./src/auth.ts
RIGHT:
/Users/turla/Code/myproject/src/auth/oauth-handler.ts
Recent Changes
WRONG:
- "Updated the file"
- "Fixed some issues"
RIGHT:
- "Added
validateTokenExpiry()function to/Users/turla/Code/myproject/src/auth/token-storage.ts(lines 42-67) that checks JWT expiration. ModifiedrefreshToken()to call this before attempting refresh."
Immediate Next Step
WRONG:
- "Continue working"
- "Finish the implementation"
RIGHT:
- "Run
npm test -- --testPathPattern=oauth-handlerto verify the newvalidateTokenExpiry()function works with expired tokens"
Validation Checklist
Section 1
- Goal is 2+ sentences with Definition of Done
- Status is exactly: Completed, In Progress, or Blocked
- Next step is a specific command or edit
Section 2
- All file paths are absolute (start with
/) - Recent changes reference specific function/method names
Section 3
- Discarded approaches explain WHY they were discarded
- No empty sections (use "None stated this session" if applicable)
Section 4
- Starts with "When resuming, the first thing to do is..."
- Contains a specific, actionable instruction
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
rewind
List and restore to a named checkpoint from the current session. Use when the agent has gone off track and you need to recover to a previous good state.
consume-content
Produce a faithful content-snapshot of any source material (article, report, PDF, advisory) with verbatim quotes, structural transparency, and labeled editorial.
deep-research
Multi-model deep research with comparative assessment (OpenAI + Perplexity + Gemini). Queries 3 deep research providers in parallel and produces a comparative synthesis.
reckoning
Analyze a project's MASTER_PLAN.md to assess coherence, evolution trajectory, and intent alignment. Modes: default (full analysis), compare (delta between reckonings), operationalize (convert findings to actionable work via /decide), steer (strategic brainstorming grounded in findings).
decide
Generate an interactive decision configurator from research or plan analysis. Presents options as explorable cards with trade-offs, costs, and filtering. Integrates with Planner to collect DEC-ID decisions.
diagnose
Validate hook integrity, state file consistency, and system health for the ~/.claude configuration.
Didn't find tool you were looking for?