Results for “context-poisoning”

Diagnose and mitigate context degradation patterns including lost-in-middle failures, context poisoning, distraction, confusion, and clash in AI agent systems.

Place trusted contextual information into prompts or agent state using explicit boundaries, provenance, and templates. Use when relevant context has already been selected and must be inserted safely; use context-retrieval to find it or context-optimization to choose and order it.

Build the smallest, highest-signal context package for an AI coding task — goal, constraints, repo facts, boundaries, and a verification plan. Load when prompts are underspecified, the agent is missing key files or decisions, the user says "use the right context", "here's the repo", or when work is drifting due to missing constraints. Also triggers on "context engineering", "gather context", "what do you need from me", "before you start". Not for cross-session continuity (use memory-startup/memory-recall).

Diagnose whether an AI workflow suffers from context stuffing or benefits from context engineering, and apply structured techniques to improve reliability.

Extends effective context capacity through strategic compression, masking, caching, and partitioning techniques.

Optimize a complete candidate context package by deduplicating, filtering, ordering, and allocating its token budget. Use when retrieved or assembled material is noisy or exceeds the useful context budget; use context-ranking for scoring chunks and context-compression for shrinking selected content.

Context Management

Optimizes long-running agent sessions with structured context compression, summarization, and durable handoff summaries that preserve decisions, files, risks, and next actions.

Use when context window is filling up, sessions are getting slow, or you need to preserve state across sessions. For context compression, state snapshots, search-first discipline, and cross-session memory management.

Optimizes agent context setup by structuring rules, specs, source files, error output, and conversation history to improve output quality.

Breaks down current state, context, user input, and data into a structured assessment before taking action.

Detect poisoned training data and backdoored models across the ML pipeline using statistical analysis, activation clustering, and spectral signatures.

Execute and analyze AI Data Poisoning attacks. By subtly injecting malicious or targeted misinformation into an LLM's training or fine-tuning dataset, an attacker can covertly manipulate the model's future outputs, implant backdoors, or enforce biases without altering the model architecture.

Rank an existing set of context chunks by relevance, diversity, freshness, and utility. Use when retrieval has already produced candidates that must be scored or reranked; use context-retrieval when the source corpus still needs to be searched.

Context window monitoring and budget management. Keeps orchestrator at 15-30% context usage while subagents get full 200k tokens. Provides warnings at thresholds, context-aware summarization triggers, and wave-level budget planning.

Probe RAG applications for prompt injection via poisoned retrieved context and embedding manipulation.

Loads and manages company context for C-suite advisor sessions, detecting stale context, enriching it during conversations, and enforcing privacy rules before external API calls.

Use this skill for context gathering, file triage, source maps, assumptions, task framing, prompt hygiene. Trigger when the task involves programming work related to Context Engineering, production implementation, audits, debugging, strategy, or validation.

Review and prune the project context — archive completed work, summarize stale entries, and propose disabling unused skills.

Optimize context usage for the session by reducing bloat, improving token efficiency, and focusing on relevant areas. Use when context is growing large, responses are slow, or switching between unrelated tasks. Covers context analysis, compacting, targeted pruning, MCP-first strategies, and focused scoping by area.

Exploit web cache mechanisms to serve malicious content to other users by poisoning cached responses through unkeyed headers and parameters during authorized security tests.

Capture and maintain a foundational marketing context document covering positioning, messaging, brand voice, target audience, and competitive landscape for consistent use across marketing skills.

Compress selected context to a target token budget while preserving decisions, evidence, constraints, and unresolved questions. Use when relevant material is already selected but too long; use context-optimization when selection, deduplication, and ordering are also required.

Loads and manages company context for all C-suite advisor skills. Reads ~/.claude/company-context.md, detects stale context (>90 days), enriches context during conversations, and enforces privacy/anonymization rules before external API calls.

Build students' capacity to place historical documents in their temporal and social context. Use when students read sources without considering what was happening at the time, or know the context but don't deploy it.

Retrieve relevant information from a knowledge base using semantic, keyword, or hybrid search to ground a query. Use when the task starts with a corpus or index that must be searched; use context-ranking when candidate chunks already exist and only need ordering.

`task-agent`/`review-agent`: use when image layers, build context, runtime user, secrets, health checks, shutdown, or provenance change; skip when container behavior is unaffected.

Automates multi-source enrichment of IPs, domains, URLs, and file hashes using VirusTotal, AbuseIPDB, Shodan, GreyNoise, URLScan.io, and MISP to provide contextual risk scoring and disposition recommendations for SOC analysts.

Progressively refines context retrieval in multi-agent workflows to solve the subagent context problem.

Builds a compact secret-safe repository pack for AI handoff, review, or large analysis; prefers Repomix when available.

Use when a project's durable context needs to be seeded or refreshed — creating or updating docs/product/context.md (overview + digest + source index + glossary + Know/Don't-Know ledger + design context + technical context) in the target repo; when a project has no context layer yet (Stage-0 setup); when domain terminology is still forming or has drifted; after a meeting, new evidence batch, or scope change that outdates the current context. Also use to ingest new docs/product/input/ batches into the digest. Do not use for repo scaffolding or CI setup, for writing the discovery plan, PRD, or epics (use discovery-plan, prd-writer, epics), or for creating or revising Jobs to be Done — this skill records finished JTBDs by id and marks missing ones TBD; jobs-to-be-done owns minting them.

Select evidence-bearing context for a downstream consumer after work when claims, artifacts, validation, unresolved decisions, or omissions must cross a boundary.

Manage and compress conversation context in long sessions. Detect when context is growing large, summarize completed work phases, archive old findings while preserving key decisions. Prevents context degradation.

Selects a minimal, non-redundant set of facts, artifacts, skills, and references for a working decision when context is stale, uncertain, or over budget.

Use the context-grooming skill to prune and compact the current project context.