AgentSkillsCN

Skill Activation

技能激活

SKILL.md

Skill Activation

Meta-cognitive skill for automatic capability discovery. Self-triggers when uncertain.

Purpose

Before answering ANY task request, Alex automatically consults the action-keyword index below. This is an internal pre-processing step, not a user-triggered action.

Auto-Trigger Conditions

This skill activates automatically when Alex:

  1. Is about to suggest manual steps for a task
  2. Is uncertain whether a capability exists
  3. Is formulating a response that includes "you can...", "try...", "manually..."
  4. Encounters an action verb (convert, create, generate, build, debug, etc.)

NOT triggered by user phrases — this is internal metacognition.

Action-Keyword Index

Path Pattern: .github/skills/{skill-name}/SKILL.md

Search this index when processing any task request:

SkillAction Keywords
academic-researchwrite thesis, literature review, cite sources, research paper, dissertation
ai-agent-designdesign agent, react pattern, multi-agent, tool use, agent architecture
airs-appropriate-relianceairs survey, measure adoption, psychometric scale, utaut, ai readiness
alex-effort-estimationestimate effort, how long, task duration, ai time, planning
anti-hallucinationprevent hallucination, verify claim, admit uncertainty, fact check, don't know
api-designdesign api, rest endpoints, openapi, http status, api versioning
appropriate-reliancecalibrate trust, when to challenge, confidence level, human-ai collaboration
architecture-auditaudit project, consistency check, version drift, fact inventory, pre-release audit
architecture-healthcheck health, validate synapses, neural maintenance, health report, dream
architecture-refinementrefine architecture, document pattern, consolidate files, update tracker
ascii-art-alignmentalign ascii, box drawing, unicode boxes, ascii diagram, fix alignment
awarenessself-correct, detect error, temporal uncertainty, version caveat, red flag phrase
bootstrap-learninglearn topic, bootstrap learning, teach me, build knowledge, skill creation
business-analysiswrite brd, gather requirements, stakeholder analysis, use cases, process map
change-managementadkar, manage change, stakeholder engagement, adoption strategy, transition plan
chat-participant-patternschat api, vscode participant, chat handler, stream response, copilot extension
code-reviewreview code, review pr, feedback comment, blocking issue, approve merge
cognitive-loadreduce complexity, chunk information, simplify explanation, progressive disclosure
creative-writingwrite story, character arc, plot structure, dialogue, narrative
debugging-patternsdebug error, binary search debug, read stack trace, git bisect, isolate bug
error-recovery-patternsretry logic, circuit breaker, fallback pattern, rollback, error handling
⭐ fabric-notebook-publishpush to fabric, sync notebook, fabric git, ado worktree, notebook changelog
⭐ gamma-presentationscreate presentation, slide deck, gamma api, pitch deck, generate slides
git-workflowgit commit, git recovery, undo commit, restore file, branch strategy
global-knowledgesearch knowledge, cross-project, find pattern, save insight, reuse solution
grant-writingwrite grant, nsf proposal, nih application, specific aims, funding
graphic-designvisual hierarchy, layout grid, typography, color palette, composition
heir-curationcurate heir, package extension, exclude files, clean payload, heir audit
image-handlingconvert svg, svg to png, resize image, sharp-cli, image optimization
incident-responsehandle incident, severity triage, outage response, incident timeline, on-call
infrastructure-as-codeterraform, bicep, provision infrastructure, iac, cloudformation
knowledge-synthesissynthesize knowledge, abstract pattern, promote insight, cross-project learning
learning-psychologyteach naturally, zone proximal, adaptive learning, learning partnership
lint-clean-markdownfix markdown lint, blank lines, md032, clean markdown, lint rules
llm-model-selectionchoose model, opus vs sonnet, claude model, cost optimization, model tier
localizationtranslate, i18n, localize app, language detection, rtl support
m365-agent-debuggingdebug m365 agent, declarative agent, manifest validation, copilot agent
markdown-mermaidcreate diagram, mermaid syntax, flowchart, sequence diagram, visualize
master-alex-auditfull audit, heir sync, 22-point check, pre-release, security audit
mcp-developmentbuild mcp server, mcp tools, model context protocol, mcp client
meditationmeditate, consolidate knowledge, reflect session, memory integration
meditation-facilitationguide meditation, four r's, deep dive, self-actualize
⭐ microsoft-fabricfabric api, medallion architecture, lakehouse, unity catalog, fabric governance
microsoft-sfisecurity review, stride threat, secure by design, sfi compliance, threat model
pii-privacy-regulationsgdpr compliance, pii handling, data protection, privacy audit, consent
post-mortemrun post-mortem, blameless review, incident analysis, action items, 5 whys
practitioner-researchresearch methodology, case study, evidence collection, academic writing
privacy-responsible-airesponsible ai, ethical ai, bias detection, fairness, data minimization
project-deploymentdeploy project, npm publish, pypi upload, release package, cargo publish
project-managementmanage project, rapid iteration, session workflow, planning document
project-scaffoldingscaffold project, create readme, init repo, hero banner, new project
prompt-engineeringwrite prompt, chain of thought, few-shot, prompt template, system prompt
rag-architecturebuild rag, vector search, embeddings, retrieval augmented, chunking
refactoring-patternsrefactor code, extract function, code smell, safe refactor, inline
release-preflightpreflight check, version sync, pre-release, bump version, validate release
release-processpublish extension, vsce publish, marketplace, pat token, vsix
research-project-scaffoldscaffold research, literature matrix, research plan, data dictionary
root-cause-analysisfind root cause, 5 whys, cause category, timeline analysis, fix prevent
rubber-duck-debuggingexplain problem, thinking partner, stuck debugging, talk through
security-reviewsecurity audit, owasp check, vulnerability scan, auth review, stride
self-actualizationself-actualize, deep assessment, architecture review, comprehensive check
skill-catalog-generatorshow skills, skill catalog, skill network, learning progress
socratic-questioningask questions, discover answer, probe assumption, socratic method
svg-graphicscreate svg, svg banner, svg icon, dark mode svg, scalable graphic
teams-app-patternsteams app, teams manifest, adaptive card, teams bot, teams sso
testing-strategieswrite tests, unit test, test coverage, mock dependencies, tdd
⭐ text-to-speechread aloud, tts, edge tts, speak text, voice synthesis, audio playback
vscode-environmentsetup vscode, workspace settings, launch.json, extensions.json
vscode-extension-patternswebview pattern, extension api, tree provider, vscode extension
work-life-balancedetect burnout, take break, late night, sustainable productivity
writing-publicationacademic paper, technical writing, publication strategy, cars model

Protocol

Step 1: Intercept Response Formation

Before generating any task-oriented response:

  • PAUSE internal response generation
  • Extract action + object from user request
  • Proceed to Step 2

Step 2: Search Action-Keyword Index

Scan the table above:

  • Match extracted keywords against skill triggers
  • Identify applicable skills
  • If match found → load skill from .github/skills/{name}/SKILL.md, execute
  • If no match → proceed with best available approach

Step 3: Execute or Acknowledge

ResultAction
Skill foundExecute using skill knowledge
No skill, but can doProceed, note potential new skill
Cannot doAcknowledge limitation honestly

Self-Correction Protocol

If Alex catches itself mid-response suggesting manual work:

  1. Stop
  2. Internal: "Wait — check skills first"
  3. Search action-keyword index above
  4. If skill exists: "Actually, I can do this." → Execute
  5. If no skill: Continue with original response

Failure Mode: The SVG→PNG Incident

What happened: User asked to convert SVG to PNG. Alex suggested manual browser screenshot instead of using image-handling skill with sharp-cli.

Root cause: Failed to consult action-keyword index before responding.

Prevention: This skill now auto-triggers on ANY action request.

Synapses

  • .github/skills/awareness/SKILL.md — WHEN: detecting own uncertainty | YIELDS: red flag phrases, self-correction patterns
  • .github/skills/anti-hallucination/SKILL.md — WHEN: uncertain if skill exists | YIELDS: verification protocol, admit-uncertainty
  • .github/skills/cognitive-load/SKILL.md — WHEN: multiple skill matches | YIELDS: chunking strategies, decision simplification