npx skills add ...
npx skills add neolabhq/context-engineering-kit --skill decay
Manage evidence freshness by identifying stale decisions and providing governance actions
npx skills add neolabhq/context-engineering-kit --skill decay
Manages evidence freshness by identifying stale decisions and providing governance actions. Implements FPF B.3.4 (Evidence Decay).
Key principle: Evidence is perishable. Decisions built on expired evidence carry hidden risk.
Every piece of evidence has a valid_until date. A benchmark from 6 months ago may no longer reflect current system performance. A security audit from before a major dependency update doesn't account for new vulnerabilities.
When evidence expires, the decision it supports becomes questionable - not necessarily wrong, just unverified.
Waiving = "I know this evidence is stale, I accept the risk temporarily."
Use it when:
A waiver is NOT ignoring the problem - it's explicitly documenting that you know about the risk and accept it until a specific date.
| Situation | Action | What it does |
|---|---|---|
| Evidence is old but decision is still good | Refresh | Re-run the test, get fresh evidence |
| Decision is obsolete, needs rethinking | Deprecate | Downgrade hypothesis, restart evaluation |
| Accept risk temporarily | Waive | Record the risk acceptance with deadline |
.fpf/evidence/valid_until from frontmatterBased on user response, perform one of:
User: "Refresh the redis caching evidence"
.fpf/knowledge/L2/User: "Deprecate the auth module decision"
User: "Waive the benchmark until February"
You don't need to memorize evidence IDs. Just describe what you want.
A hypothesis is STALE if any of its evidence is expired (and not waived).
This is the Weakest Link (WLNK) principle: reliability = min(all evidence). One stale piece makes the whole decision questionable.
All actions are logged:
| Action | What's Recorded |
|---|---|
| Deprecate | from_layer, to_layer, reason, date |
| Waive | evidence_id, until_date, rationale, date |
Files created in .fpf/evidence/:
deprecate-{hypothesis}-{date}.mdwaiver-{evidence}-{date}.md# In .fpf/evidence/deprecate-auth-module-2025-01-15.md
---
id: deprecate-auth-module-2025-01-15
hypothesis_id: auth-module
action: deprecate
from_layer: L2
to_layer: L1
created: 2025-01-15T10:00:00Z
---
# Deprecation: auth-module
**Reason**: Evidence expired, technology landscape changed
**Next Steps**: Run `/fpf:propose-hypotheses` to explore alternativesmv .fpf/knowledge/L2/auth-module.md .fpf/knowledge/L1/auth-module.md# In .fpf/evidence/waiver-benchmark-2025-01-15.md
---
id: waiver-benchmark-2025-01-15
evidence_id: ev-benchmark-2024-06-15
waived_until: 2025-02-01
created: 2025-01-15T10:00:00Z
---
# Waiver: ev-benchmark-2024-06-15
**Evidence**: ev-benchmark-2024-06-15
**Hypothesis**: redis-caching
**Waived Until**: 2025-02-01
**Rationale**: Migration pending, will re-run after completion
**Accepted By**: User
**Created**: 2025-01-15
**WARNING**: This evidence returns to EXPIRED status after 2025-02-01.User: /fpf:decay
Agent shows report with stale evidence
User: Waive the benchmark until February, we'll re-run it after the migration.
Agent: Creating waiver for ev-benchmark-2024-06-15 until 2025-02-01.
Rationale: "Re-run after migration"
[Creates .fpf/evidence/waiver-benchmark-2025-01-15.md]
User: The vendor API is being discontinued. Deprecate that decision.
Agent: Deprecating hypothesis-vendor-api from L2 to L1.
[Moves file, creates deprecation record]
Next step: Run /fpf:propose-hypotheses to explore alternatives./fpf:decay # See what's stale
# For each stale item: refresh, deprecate, or waive/fpf:decay # Check for stale decisions
# Either refresh evidence or explicitly waive with documented rationale
# Waiver rationales become part of release documentation# Dependency update, API change, security advisory...
/fpf:decay # See what's affected
# Deprecate obsolete decisions
# Start new hypothesis cycle for replacements