From agent-almanac
Applies ant colony optimization and foraging theory to search large solution spaces, balancing exploration of new approaches with exploitation of known good ones.
npx claudepluginhub pjt222/agent-almanacThis skill uses the workspace's default tool permissions.
---
Explores solution spaces using ant colony optimization: deploys scout hypotheses, reinforces promising approaches, abandons dead-ends. For multi-approach problems, unclear debugging, codebase searches.
Conducts deep web research with parallel agents, multi-wave exploration for gaps, and structured synthesis. Activates for investigating topics, comparing options, best practices, or comprehensive web info.
Orchestrates research workflows from question definition to evidence-based findings documentation for technical, requirements, literature, and codebase topics.
Share bugs, ideas, or general feedback.
Apply foraging theory and ant colony optimization to systematically search for, evaluate, and exploit distributed resources — balancing exploration of unknown territory with exploitation of known yields.
coordinate-swarm with specific resource-discovery patternsCharacterize the resource environment to select appropriate foraging strategy.
coordinate-swarm for signal design)Expected: A characterized foraging landscape with resource distribution type, information availability, and cost structure. This determines which foraging model to apply.
On failure: If the landscape is completely unknown, start with maximum exploration (all scouts, no exploitation) for a fixed time budget to build an initial map. Switch to the appropriate model once the landscape character becomes clear.
Send exploratory agents into the search space with instructions to mark what they find.
Expected: Scouts deployed across the search space, depositing trail signals proportional to resource quality. The initial map of the landscape begins to emerge from scout reports.
On failure: If scouts find nothing in the initial sweep, either the scout percentage is too low (increase to 50%), the search pattern is wrong (switch from random walk to Levy flight for patchy resources), or the quality assessment is miscalibrated (lower the detection threshold).
Create positive feedback loops that amplify successful paths and let unsuccessful ones fade.
Trail Reinforcement Dynamics:
┌─────────────────────────────────────────────────────────────────────┐
│ │
│ Strong trail ──→ More foragers ──→ If good: reinforce ──→ EXPLOIT │
│ ↑ │ │
│ │ If bad: no reinforce │ │
│ │ │ │ │
│ │ ↓ │ │
│ Decay ←── Weak trail ←── Fewer foragers ←── Trail fades │ │
│ │ │ │
│ ↓ │ │
│ No trail ──→ Scouts explore ──→ New discovery ──→ New trail ↗ │
│ │
└─────────────────────────────────────────────────────────────────────┘
Expected: A self-regulating feedback loop where good resources attract increasing attention and poor resources are naturally abandoned. The system balances exploitation and exploration through trail dynamics alone.
On failure: If all foragers converge on a single trail (premature convergence), the decay rate is too slow or the saturation cap is too high. Increase decay, lower the cap, or introduce random exploration mandates (e.g., 10% of foragers always ignore trails). If trails fade too fast and nothing gets exploited, reduce the decay rate.
Monitor resource yields to know when to shift from exploitation back to exploration.
Expected: The foraging swarm naturally shifts between exploitation phases (concentrated on known-good sites) and exploration phases (scouts dispersed), driven by yield monitoring rather than arbitrary schedules.
On failure: If the swarm stays on depleted sites too long, the marginal value threshold is set too low or the travel cost estimate is too high. Recalibrate by comparing actual yield rates. If the swarm abandons good sites too early, the threshold is too sensitive — add a smoothing window to the yield measurement.
Select and switch between foraging strategies based on environmental feedback.
Expected: A foraging system that adapts its exploration-exploitation balance to the current environment, maintaining effectiveness as conditions change.
On failure: If strategy adaptation itself becomes unstable (oscillating between exploration and exploitation), add damping: require the mismatch signal to persist for N time units before triggering a strategy shift. If no strategy seems to work, reassess the landscape characterization from Step 1 — the resource distribution may be more complex than initially assumed.
coordinate-swarm — foundational coordination patterns that underpin foraging signal designbuild-consensus — used when the swarm must collectively agree on which resource patches to prioritizescale-colony — scaling foraging operations when the resource landscape or swarm size growsassess-form — morphic skill for evaluating the current state of a system, complementary to landscape assessmentconfigure-alerting-rules — alerting patterns applicable to diminishing returns detectionplan-capacity — capacity planning shares the explore-exploit framing with foraging theoryforage-solutions — AI self-application variant; maps ant colony foraging to single-agent solution exploration with scout hypotheses and trail reinforcement