self-improvement Skill | Agent Skills

<skill> <name>self-improvement</name> <description> Institutionalizes learning by capturing friction, proposing experiments, and measuring impact. Agents must use Gaia improvement tools exclusively—no temporary files, no informal notes. Small wins compound into significant velocity gains. </description> <philosophy> <principle name="Tool Mandate">MUST use Gaia improvement tools (log_frustration, propose_improvement, verify_impact). No docs/, no scratch files, no TODOs.</principle> <principle name="Atomic Changes">One friction point → one proposed fix → one measurable outcome. Never bundle unrelated improvements.</principle> <principle name="Close the Loop">Every logged frustration requires a follow-up within 5 sessions to verify if the change helped or needs iteration.</principle> <principle name="Blame Process, Not People">Focus on systemic issues (tooling, docs, patterns) rather than individual mistakes.</principle> </philosophy> <friction-detection> <trigger name="Repetition">Same issue encountered twice in one session or three times across sessions.</trigger> <trigger name="Workarounds">Manual steps required to complete what should be automated or streamlined.</trigger> <trigger name="Context Loss">Information had to be re-discovered because it wasn't captured or was hard to find.</trigger> <trigger name="Tooling Gaps">Missing capability forced fallback to slower or less reliable approach.</trigger> <trigger name="Unclear Guidance">Ambiguity in skill, agent, or doc caused decision paralysis or wrong path.</trigger> <severity-threshold>Log if issue cost >5 minutes or will recur. Skip trivial one-offs.</severity-threshold> </friction-detection> <logging-protocol> <step number="1">Identify the friction using triggers above.</step> <step number="2">Log via Gaia tool: category (tooling|process|docs|skill|agent), description, impact estimate.</step> <step number="3">Propose concrete fix: what changes, who owns it, success criteria, effort estimate (S/M/L).</step> <step number="4">Tag with priority: P1 (blocks work), P2 (slows work), P3 (annoyance).</step> <format> Category: [tooling|process|docs|skill|agent] Friction: [What happened, why it hurt] Impact: [Time lost, risk introduced, or frequency] Proposed Fix: [Specific change] Owner: [architect|developer|orchestrator|self] Success Criteria: [How we know it worked] Effort: [S: <1h, M: 1-4h, L: >4h] Priority: [P1|P2|P3] </format> </logging-protocol> <verification-protocol> <timing>Review logged items every 5 sessions or when similar friction recurs.</timing> <questions> <question>Did the proposed fix get implemented?</question> <question>Did it reduce or eliminate the friction?</question> <question>Any unintended side effects?</question> <question>Should we iterate, close, or escalate?</question> </questions> <outcomes> <outcome name="Verified">Fix worked, close the loop, optionally share as pattern.</outcome> <outcome name="Partial">Some improvement, propose iteration.</outcome> <outcome name="Failed">No improvement, analyze root cause, try different approach.</outcome> <outcome name="Superseded">Problem no longer relevant, close without action.</outcome> </outcomes> </verification-protocol> <escalation-rules> <rule>P1 items: escalate to orchestrator immediately if not resolved in current session.</rule> <rule>Recurring P2 items (3+ occurrences): escalate to architect for systemic fix.</rule> <rule>Skill/agent gaps: propose skill update or new skill via skill-maturity process.</rule> <rule>Cross-agent friction: orchestrator coordinates resolution across agents.</rule> </escalation-rules> <anti-patterns> <anti-pattern name="Silent Struggling">Working around issues without logging them. Always log—even if you fix it yourself.</anti-pattern> <anti-pattern name="Vague Proposals">"Improve the docs" is not actionable. Specify which doc, what change, why.</anti-pattern> <anti-pattern name="Over-Engineering">Proposing L-effort fixes for P3 annoyances. Match effort to impact.</anti-pattern> <anti-pattern name="Orphaned Logs">Logging without follow-up. The loop must close.</anti-pattern> <anti-pattern name="File-Based Tracking">Creating TODO.md, NOTES.md, or similar. Use Gaia tools only.</anti-pattern> </anti-patterns> <success-metrics> <metric>Friction logged within session of occurrence (no backlog drift).</metric> <metric>80%+ of logged items have verified outcomes within 10 sessions.</metric> <metric>Recurring friction decreases over time (measured by duplicate log entries).</metric> <metric>Zero temporary tracking files in repository.</metric> <metric>Improvement proposals are specific, actionable, and appropriately sized.</metric> </success-metrics> <integration> <with-skill name="memory">Store successful fixes as patterns for future recall.</with-skill> <with-skill name="skill-maturity">Skill gaps identified here feed into maturity assessments.</with-skill> <with-agent name="orchestrator">P1 escalations route through orchestrator for prioritization.</with-agent> <with-agent name="architect">Systemic process/doc issues route to architect for resolution.</with-agent> </integration> </skill>

Agent Skills: self-improvement

Install this agent skill to your local

Skill Files