hooks metrics prints 'undefined' in text while --json returns 45 patterns / 6 memories; and 'Routing accuracy' counts post-edit calls, not routes — hooks route records nothing

Summary

Two independent problems in hooks metrics:

1. The text output prints undefined for Patterns / Memories / Error patterns, while --json returns the correct values from the same invocation. The data is present; the text renderer reads fields that do not exist.
2. "Routing accuracy" does not measure routing. hooks route never records a route — the counter stays at 0 no matter how many routing decisions are made. The only writer is hooks post-edit, so the figure reported as routing accuracy is really "the fraction of edits the caller passed --success for".

Version: agentic-flow 2.1.2, Node v22.23.0, macOS 15.

1. undefined in text, correct values in JSON

Same command, same moment, same store:

$ npx agentic-flow hooks metrics

📚 Learning:
   Patterns: undefined
   Memories: undefined
   Error patterns: undefined

💚 Health: NEEDS-DATA

$ npx agentic-flow hooks metrics --json
learning.patternsLearned = 45
learning.memoriesStored  = 6
memory.total             = 6
memory.byType.project    = 6
errors.total             = 0
topPatterns              = [5 items]

Both agree with what is on disk:

.agentic-flow/intelligence.json → patterns 45, memories 6

So the JSON path reads the store correctly and the text path prints undefined — it appears to be looking for patterns / memories rather than patternsLearned / memoriesStored. topPatterns (5 entries of real data) has no text representation at all.

This matters because the text output is the default. A user running hooks metrics after a successful pretrain sees undefined and NEEDS-DATA and concludes the store is empty, when it holds 45 patterns.

2. hooks route does not record routes

metrics() { python3 -c "import json;print(json.load(open('.agentic-flow/intelligence.json'))['metrics'])"; }

metrics                                             # {'totalRoutes': 0, 'successfulRoutes': 0, 'routingHistory': []}
npx agentic-flow hooks route "fix a bug in the parser"
metrics                                             # {'totalRoutes': 0, ...}   ← unchanged

npx agentic-flow hooks post-edit src/roster.cjs --success --agent coder
metrics
# {'totalRoutes': 1, 'successfulRoutes': 1,
#  'routingHistory': [{'task': 'edit:src/roster.cjs', 'agent': 'coder', 'success': True}]}

hooks route produced a recommendation and recorded nothing. The single history entry comes from post-edit, and its shape shows what is actually being counted:

• task: "edit:src/roster.cjs" — a file edit, not a routing request
• agent: "coder" — supplied by the caller via --agent, not chosen by the router
• success: true — supplied by the caller via --success

None of those three fields is a routing decision or an evaluation of one. So Routing accuracy: 100% after a successful edit means "the caller said the edit worked", and a user could drive that number to 100% or 0% by passing --success or --fail, without the router ever being consulted.

Consequently the Total routes: 0 shown above is accurate but misleading — routes were made, they are simply never counted.

Suggested fix

1. Fix the text renderer to read patternsLearned / memoriesStored (and surface topPatterns, which is currently JSON-only).
2. Record routing decisions in hooks route, so totalRoutes counts routes.
3. Either separate the two counters — routing decisions vs edit outcomes — or rename the metric to reflect what it measures. As it stands, "Routing accuracy" is a caller-controlled edit success rate.

A test asserting that the text and --json outputs report the same numbers for the same store would catch item 1, and asserting totalRoutes increments after hooks route would catch item 2.

Related

• #186 — hooks route: confidence fixed at 95.0%, factors NaN, alternatives outrank the recommendation. It also, per this issue, records nothing.
• #198 — hooks explain: [object Object] in text output while --json carries the real structured data. Same class of text-renderer defect.
• #185, #199, #200 — other hooks subcommands whose reported numbers disagree with what is on disk.

Start with the hooks metrics, hooks route, and hooks post-edit entry points, comparing text and --json output against .agentic-flow/intelligence.json; src/roster.cjs is the example edit path. Add or update tests for matching text/JSON learning values and for route counts changing after hooks route. Done means text shows the stored metrics and routing metrics represent routing decisions rather than caller-reported edit outcomes.