Slash Command

/instinct-outcome

Records good/bad/neutral outcome for an instinct ID to adjust its confidence score, shows updated score and change. Also supports listing instincts.

Node

Bash

developer-tools

Popularity

Stars

Invocation

How this command is triggered — by the user, by Claude, or both

Slash command

/clarc:instinct-outcome

Model invocable

No pre-commands

Context Preview

The summary Claude sees in its command listing — used to decide when to auto-load this command

# Instinct Outcome Command

Record an outcome for a specific instinct. This updates its confidence score and outcome counters.

## Usage



## Implementation

Run the outcome tracker script:



Or if `CLAUDE_PLUGIN_ROOT` is not set (manual installation):



To list available instinct IDs:



## Confidence Adjustments

| Outcome | Confidence change |
|---------|-----------------|
| `good`  | +0.05 (max 0.95) |
| `bad`   | -0.10 (min 0.10) |
| `neutral` | no change |

## Preconditions

Before running, verify:
- `~/.claude/homunculus/` exists — if not, instruct user to run `install.sh --enable...

Command Content

91 lines · ~706 tokens

Stats

LanguageJavaScript

Stars9

MaintenanceExcellent

Last CommitApr 7, 2026

Actions

View Source View Plugin View on GitHub View README

Instinct Outcome Command

Record an outcome for a specific instinct. This updates its confidence score and outcome counters.

Usage

/instinct-outcome <id> <good|bad|neutral>
/instinct-outcome <id> good "reason why it worked"
/instinct-outcome <id> bad "reason it caused problems"
/instinct-outcome list

Implementation

Run the outcome tracker script:

node "${CLAUDE_PLUGIN_ROOT}/scripts/instinct-outcome-tracker.js" <id> <outcome> [reason]

Or if CLAUDE_PLUGIN_ROOT is not set (manual installation):

node ~/.claude/scripts/instinct-outcome-tracker.js <id> <outcome> [reason]

To list available instinct IDs:

node "${CLAUDE_PLUGIN_ROOT}/scripts/instinct-outcome-tracker.js" --list

Confidence Adjustments

Outcome	Confidence change
`good`	+0.05 (max 0.95)
`bad`	-0.10 (min 0.10)
`neutral`	no change

Preconditions

Before running, verify:

~/.claude/homunculus/ exists — if not, instruct user to run install.sh --enable-learning
The instinct ID is valid — run list subcommand if unsure of the ID
Outcome is one of good, bad, or neutral

Execution

Ask the user which instinct they want to record an outcome for (or use the list subcommand)
Validate the outcome is good, bad, or neutral
Run the tracker script with the instinct ID and outcome
Show the updated confidence score
Suggest running /instinct-report to see the updated ranking

Example Output

Recorded good outcome for "test-first-workflow"
  Confidence: 0.70 ↑ 0.75
  Reason: TDD caught a regression before it hit staging

Output Interpretation

Confidence increased: instinct is gaining evidence — will appear higher in /instinct-report
Confidence decreased: if drops below 20%, it will be flagged in /instinct-report for removal review
Run /instinct-report after recording: to see the updated ranked list with trend indicators (↑↓→)
Pattern: if the same instinct consistently gets bad outcomes, consider using /evolve to revise it

Notes

Outcomes are logged to ~/.claude/homunculus/outcomes.jsonl for audit
last_used is updated to today on every outcome recording
usage_count increments on every outcome recording
Use /instinct-status to see current confidence levels
Use /instinct-report to see ranked list with trends

After This

/evolve — trigger instinct evolution when enough outcomes are collected
/instinct-report — see the updated ranked instinct list with trend indicators
/instinct-status — check current confidence levels across all instincts