**Status:** Proposed (extends existing backlog items #1 and #3) **Priority:** High **Affects:** /spark, /explore, /pulse, fascination index **Inspired by:** GSD's gsd-phase-researcher, Agent Patterns bounded pipeline, researchfun FUN protocol, Compound Engineering parallel research agents

This skill should be used when the user says 'autopsy', 'shelve', 'done with this', 'extract value', or wants to close out a dead or shelved project. Extracts fascinations, writes a composting report, and feeds insights back into /spark.

This skill should be used when the user needs live research, source verification, fact-checking, current comparisons, or updates to an existing research note. Responds to "look this up", "what's the latest on X", "compare X vs Y", "is this actually true", "fact check this", "update my research notes", or any question requiring evidence-grounded answers with strong next-step branching for deeper investigation.

This skill should be used when the user says 'contract', 'derive', 'extract contracts', 'freeze', 'invariants', 'nfr', or wants to lock down requirements from a converged prototype. Derives contracts, invariants, and NFR targets, freezes on approval.

This skill should be used when the user wants to pressure-test a feature, flow, or product idea by generating emotionally concrete future failure artifacts such as support tickets, angry emails, user complaints, one-star reviews, or postmortem fragments. Responds to "what could go wrong", "stress test this", "write a pre-mortem", "show me how this fails", "what would the support tickets look like", "pressure test this flow", or any request to make abstract risk viscerally real before or during design.

This skill should be used when the user says 'fascination', 'what am I into', 'patterns', 'show my interests', or wants to browse their fascination index. Displays themes, intensities, and cross-fascination connections in list or Mermaid map mode.

This skill should be used when the user wants to define a project's identity, build a lorebook, articulate what a product believes and refuses to become, or capture the implicit rules a project lives by. Responds to "what is this project trying to be", "the product is drifting", "write down the project values", "give this project a spine", "build a lorebook", or any moment when naming, tone, scope, or product identity feel undefined or are starting to drift.

This skill should be used at session start, when entering a new project, or when the user says 'wake up', 'load memory', 'boot', 'memory-boot'. Assembles identity, procedural knowledge, project context, and recent session state into a boot packet from the agent's vault.

This skill should be used when the agent needs to save something to memory — observations, lessons, decisions, reflections, dreams, or past memories from external sources. Responds to 'remember this', 'capture', 'save to memory', 'I learned', 'note this down', or any request to persist information across sessions.

This skill should be used at end of session, when wrapping up work, or when the user says 'close session', 'end session', 'session summary', 'sign off'. Writes session summary, enforces inbox aging, evaluates pending promotions, and updates active context.

This skill should be used when the agent needs a reminder of what the memory system can do, what commands are available, or how the memory system works. Responds to 'memory help', 'what memory commands exist', 'how does my memory work', 'what can I remember', or any confusion about the memory system's capabilities. Also invoked by memory-boot after first boot to teach the agent the system.

This skill should be used when the agent needs to evaluate or promote memory records through the trust pipeline. Responds to 'promote', 'evaluate inbox', 'elevate trust', 'promote to global', or when inbox items have accumulated and need triage. Converts inbox items to threads/lessons, elevates trust levels, promotes project knowledge to global.

This skill should be used when the user says 'narrow', 'decide', 'pick one', 'lock direction', or is ready to commit to one design direction from /explore. Synthesizes reactions into a recommendation, locks on acceptance.

This skill should be used when the user says 'what should I do next', 'where am I', 'pyro', 'status', asks about project state, or needs lifecycle navigation. Reads project state and recommends the next skill to invoke. Also handles 'pyro init', 'pyro status', and 'pyro list'.

This skill should be used when the user says 'reframe', 'bored', 'novelty', 'stuck', 'different angle on remaining work', or wants fresh energy on stale tasks. Injects novelty through creative domain lenses — each lens produces one concrete actionable move.

This skill should be used when the user says 'revive', 'old project', 'bring back', 'resurrect', or wants to evaluate an abandoned repo. Archaeological analysis with scored revival options (full revival, soul transplant, organ harvest, graceful burial).

This skill should be used when the agent needs to perform web searches, look up current information online, or gather search results from the internet. Provides privacy-respecting web search through a self-hosted SearXNG instance with structured JSON output. Responds to "search the web for", "look this up online", "find recent articles about", "what does Google say about", "web search", or any task requiring live internet search results. No API keys or external accounts needed.

This skill should be used when the user says 'surface', 'prototype', 'build it', 'show me', or wants a working interactive prototype from their locked direction. Iterates on behavioral critique until convergence.