From fpl-hsmem
Full Hindsight integration diagnostic. Checks Docker, API, bank state, hook state files, and config resolution. Use when the user reports "memory isn't working", "I don't see recall happening", "retain seems broken", or wants a comprehensive health report before changing settings.
How this skill is triggered — by the user, by Claude, or both
Slash command
/fpl-hsmem:diagnoseThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Run a structured 5-step diagnostic. Each step produces a status line.
Run a structured 5-step diagnostic. Each step produces a status line. At the end, summarize as "all green" / list of issues + remediation.
curl -fsS http://localhost:8888/health
{"status":"healthy",...} → server is updocker logs hindsight)memory_get_current_bank.mcp.json env, or project name derivation
(git rev-parse --git-common-dir in the project)memory_status for memory count and document countmental_model_list for living pagesretainEveryNTurns is too highdocument_ingest runs yet (normal for new bank)mental-model skill)ls -la ~/.hindsight/state/ 2>/dev/null
Look for:
turns.json — tracks turn count per session (used by retain throttling)
retention.json — tracks message_count per session (used for compaction detection)
✓ both exist with recent mtime → hooks are running
✗ neither exists → hooks aren't firing. Check
.claude/settings.local.json for hook registration, or check if the
plugin is loaded (run claude plugin list)
Read in order, report which one provides each value:
~/.hindsight/config.json<cwd>/.mcp.json → mcpServers.hindsight.env<cwd>/.hindsight.jsonHINDSIGHT_*)Show the user the final resolved bankId and URL with their source.
Look for:
<cwd>/.hindsight-disabled fileHINDSIGHT_DISABLED=true envIf present, none of the above matters — Hindsight is disabled in this project. Tell the user and ask if they want to enable it.
Hindsight diagnostic
─────────────────────
Server: <green|red> (<details>)
Bank: <bank-id> resolved from <source>
Content: <N memories>, <M docs>, <K pages>
Hooks: <active|inactive> (state files: <found|missing>)
Opt-out: <yes|no>
<verdict: "Everything looks correct" | "Issues: ..." | "Fix path: ...">
If issues are found, propose one concrete action the user can take next, not a list of every possibility.
npx claudepluginhub forgeplan/marketplace --plugin fpl-hsmemGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.