| name | feather-fall |
| description | Feather Fall is the spell for bad situations that are already in motion. A deploy is going sideways, a service is thrashing, a queue is backing up, or an integration is failing faster than the humans can think. Instead of pretending you can teleport back to normal instantly, you slow the fall: circuit breakers, degraded modes, load shedding, safe defaults, and graceful shutdown paths. The point is not elegance. The point is buying time without multiplying damage. A good Feather Fall plan makes failure survivable and visible enough to recover from. |
| version | 1.0.0 |
| author | Wizards of the Ghosts |
| license | CC0-1.0 |
| compatibility | Hermes Agent skills system |
| metadata | {"hermes":{"tags":["spell","shipping-now","metaphorical","repair-and-recovery","recovery","repair","triage","stabilization"]}} |
Feather Fall
Turn a hard crash into a controlled descent.
What This Skill Does
Feather Fall is the spell for bad situations that are already in motion. A deploy is going sideways, a service is thrashing, a queue is backing up, or an integration is failing faster than the humans can think. Instead of pretending you can teleport back to normal instantly, you slow the fall: circuit breakers, degraded modes, load shedding, safe defaults, and graceful shutdown paths. The point is not elegance. The point is buying time without multiplying damage. A good Feather Fall plan makes failure survivable and visible enough to recover from.
In this grimoire, Feather Fall is treated as a metaphorical spell with a shipping-now delivery profile.
Canonical reference input: Feather Fall (spell).
When To Use
- A live service, deploy, or workflow is failing and needs a safer landing path immediately.
- The right answer is graceful degradation, fallback behavior, or controlled shutdown rather than business-as-usual.
- You need to define what stays up, what gets dropped, and how to keep users from falling with the system.
Prerequisites
- No extra runtime dependencies beyond Hermes Agent and the normal toolset for this session.
Procedure
- Restate the target, the success condition, and any no-touch boundaries before taking action.
- Identify the failure that is already underway, including the blast radius if nothing changes.
- Choose the descent mechanisms that reduce harm fastest: feature flags, circuit breakers, throttles, queue draining, or static fallbacks.
- Define the minimal safe experience during the fall, along with the signals that prove the descent is stabilizing.
- Return the emergency sequence, operator checks, and the criteria for either recovery or a full shutdown.
- Package the result as the deliverables below, with confidence, assumptions, and unresolved risk called out explicitly.
Deliverables
- A graceful-degradation or safe-landing sequence.
- Trigger thresholds and operator signals for entering and exiting fallback mode.
- A note on what user experience, data guarantees, or capacity are intentionally reduced during the descent.
Pitfalls / Guardrails
- Keep the metaphor anchored to a real mechanism instead of drifting into lore.
- Do not use graceful degradation to hide the incident from operators who need to respond.
- Fallback modes must protect data integrity and user safety before they protect polish or convenience.
Verification
- Check that the result includes every deliverable promised above.
