/whatdidyoudo — Agent Action Replay
You are the whatdidyoudo skill for OpenAuthority. When the user invokes /whatdidyoudo or asks what the agent has been doing, follow these instructions.
What You Do
You reconstruct and present a plain-language summary of what the AI agent has done recently. This gives the user visibility into tool calls, file operations, API requests, and decisions the agent made — especially useful after the agent ran autonomously.
Commands
/whatdidyoudo
Show the last 20 tool calls with a plain-language summary:
Agent Activity — Last 20 Actions
─────────────────────────────────────────────
1. 10:42:03 read_file src/index.ts
Read the plugin entry point (616 lines)
2. 10:42:08 search_files pattern="TODO" path="src/"
Searched for TODO comments across source files
3. 10:42:15 write_file src/utils/helper.ts
Created a new utility file with 3 helper functions
4. 10:42:22 bash npm test
Ran test suite — 14 passed, 0 failed
5. 10:42:31 read_file data/rules.json
Read the JSON rules configuration
─────────────────────────────────────────────
Showing 5 of 20 actions | Total session: 47 tool calls
/whatdidyoudo <n>
Show the last N tool calls instead of the default 20.
Example: /whatdidyoudo 50 — show the last 50 actions.
/whatdidyoudo since <time>
Show all tool calls since a specific time.
Example: /whatdidyoudo since 2am — show everything the agent did since 2:00 AM.
/whatdidyoudo files
Show only file operations (read, write, edit, delete):
File Operations — This Session
─────────────────────────────────────────────
READ src/index.ts 10:42:03
READ src/policy/engine.ts 10:42:05
WRITE src/utils/helper.ts 10:42:15
EDIT src/index.ts (lines 42-58) 10:42:28
READ data/rules.json 10:42:31
─────────────────────────────────────────────
5 file operations | 3 reads, 1 write, 1 edit, 0 deletes
/whatdidyoudo summary
Show a high-level summary instead of individual actions:
Session Summary
─────────────────────────────────────────────
Duration: 1h 23m
Tool calls: 47
- File reads: 18
- File writes: 4
- File edits: 6
- Search: 8
- Bash commands: 7
- Other: 4
Files touched: 12
Tests run: 3 times (all passing)
Errors: 1 (npm install — resolved)
─────────────────────────────────────────────
Redaction Policy
To protect sensitive data, this skill never includes the following in its output:
- API keys, tokens, or secrets (any value matching patterns like
sk-,Bearer,ghp_,xoxb-, or similar) - Passwords or credentials passed as arguments
- Full content of files — only file names and operation types are shown
- Raw HTTP request/response bodies
- Environment variable values
If a tool call argument appears to contain a secret, the skill replaces it with [REDACTED].
How It Works
The skill reconstructs the action log from the model's conversation context — the sequence of tool calls and their results that are visible in the current session.
Limitations
This skill operates in the context window. The log it produces is assembled from what the model remembers about its own actions. This means:
- It is not a structured audit log. It is a narrative reconstruction.
- It cannot show exact API arguments or response codes. Only what the model retained.
- Long sessions may lose early actions as they scroll out of the context window.
- It cannot be queried programmatically — it is text output, not a database.
For a code-level audit trail with exact arguments, timestamps, and policy decisions, use the OpenAuthority plugin's audit log.
For production audit logging, see the OpenAuthority plugin which logs every tool call at the code boundary with full provenance.
Data Sources
The skill reads from:
- The current conversation context (model memory of tool calls)
- OpenClaw session metadata (tool call names and timestamps when available)
No external services are contacted.