Free to read · no account needed

16 min · agents

Logging what matters

Goal: keep a light record of agent runs so you can debug and improve without drowning in chat logs.

Best for: repeat agent jobs (weekly research, filing, drafts).

If you cannot answer “what did it do last Tuesday?”, you do not have an agent—you have a mystery. Log the few fields that support undo and learning. You do not need enterprise observability. A simple run log works: date, goal, tools used, approvals you gave, result link, what broke. Feed patterns into failure modes and only then consider scheduling.

Why beginners skip this

Saving whole chat dumps feels thorough until you cannot find why last night’s run failed. One line beats a novel: Done? failure type? next action?

Habit this week (Asia/Shanghai): log three real runs with those three columns only—date the third line.

Step 1 — Make a run log template

One row per run in a sheet or markdown table.

Create a beginner agent run-log table with columns Date, Goal, Tools, Approvals, Result link, Issues, Next fix.
Add one sample row for a weekly competitor brief.
Example

Filled prompt

Run: MCP Drive search timed out; Done unmet; failure=timeout; next=smaller query + checklist.
ONE table row only. No transcript.

What good output looks like

  • Done boxes for “keep a light record of agent runs so you can debug” are explicit and checkable
  • Allow-list tools only; never-do lines honored
  • Stops before send/delete without human gate
  • Pass/fail: Plan→Act one→Check visible for agents-logging-what-matters

Step 2 — Log approvals, not every token

Write what you allowed (e.g. “create file in sandbox”). Skip pasting full chats—your future self will not read novels.

Summarize this chat into one run-log row:
[paste abbreviated]
Focus on approvals and outcomes.
Example

Filled prompt

After today's research-agent run: goal was competitor link pack; Done partly met; failure mode = hallucinated-URL; next change = require openable links only.
Add ONE table row (date, job, Done met?, failure mode, next change). Do not paste the chat.

What good output looks like

  • Allow-list tools only; never-do lines honored
  • Stops before send/delete without human gate
  • Pass/fail: Plan→Act one→Check visible for agents-logging-what-matters
  • Done boxes for “keep a light record of agent runs so you can debug” are explicit and checkable

Step 3 — Review weekly for patterns

Same failure twice → fix skill/instructions. Same success → keep. Ten minutes beats guessing.

From these 5 run-log rows, list patterns and 3 instruction changes:
[paste rows]
Do not suggest “log more” as the only fix.
Example

Filled prompt

Run: coding-agent on auth bug; Done met = yes; failure mode = none; next change = keep boundary card pinned.
Add ONE table row only. Five columns. No transcript.

What good output looks like

  • Exactly one table row, five columns
  • Failure mode is a single clear label (not a paragraph)
  • Next change is concrete and testable this week
  • No pasted transcript

Five fields beat forty

Date, Goal, Tools, Approvals, Result link. Everything else is optional until an incident asks for it.

Do not bookmark-only

Finish Practice now, then tick this guide on your path. Titles unread are not skills.

Close the loop

After Practice now, write one line: what you will refuse next time. That line is the habit.

Check the unknown path

What should the model do when a field is missing—null, TBD, or ask you? Write it once in Boundaries.

Red flag

Full chat dump as the only log; missing approval field; secrets or customer raw text in the log—redact and rewrite the template.

Save this card

Card: one line per run — Done? · failure type · next action. Whole-chat dumps are not logs.

1. What I shipped: ___ 2. What I refused: ___ 3. What I will never-do again: ___ (date: Asia/Shanghai)

  • [ ] Tools/writes paused
  • [ ] Refuse-line dated
  • [ ] Artifact path saved
  • [ ] Next link chosen (or stop)

Before you start

Write Done (checkboxes) and the human gate on paper before any run for “keep a light record of agent runs so you.” Allow-list tools; leave send/delete off. Guide: agents-logging-what-matters. If you only have a blank practice box, stop and fetch the real job first.

  • Goal in one line: ___
  • Allowed tools: ___
  • Forbidden: ___
  • Human approve needed for: ___

How to use this week

Create a five-column table: date · job · Done met? · failure mode · next change. After each agent-ish run this week, add one row within five minutes. on a live artifact—not a blank demo. Write one refuse-line and date the note (Asia/Shanghai). If you only skimmed agents-logging-what-matters, put 25 minutes on the calendar and redo once.

Pitfalls

PitfallFix
Saving every raw transcript foreverSummarize into rows
No result linkStore output path or doc URL
Never reviewing the logWeekly 10 minutes on calendar
Logging secrets in the Issues fieldRedact; link to private vault if needed
Example

Filled prompt

Temptation: skip the habit for agents-logging-what-matters to “save five minutes.”
Job still: keep a light record of agent runs so you
Do the smallest correct pass: Done visible · one checkable artifact · Unknowns labeled [?].
Boundaries: draft only; no silent send; no invented metrics.
Return: the artifact + the refuse-line you said aloud.

What good output looks like

  • Names the shortcut you refused (specific to agents-logging-what-matters)
  • Still delivers a checkable artifact for “keep a light record of agent runs so you”
  • Unknowns labeled; draft-only respected
  • Pass/fail: refuse-line written and dated (agents-logging-what-matters #4)

Practice now

Live card: run one agent checklist item on a real job for “Logging what matters”—artifact path on Save (Asia/Shanghai).

Today: fill a five-field run log (goal, approval, action, result, next) for one real agent run. Skip pasting the whole transcript. Date it (Asia/Shanghai).

Today card (copy) — agents-logging-what-matters: 1. Artifact I will produce: ___ 2. Pass/fail check: ___ 3. Refuse-line if tempted to skip verify/send: ___ (date: Asia/Shanghai) 4. One never-do from this guide: ___

  • [ ] Artifact path written
  • [ ] Refuse example would say "yes"
  • [ ] Dated Asia/Shanghai

One sentence takeaway

Log Done / failure type / next action—one line per run, not whole chats. This week: three dated log lines from live runs (Asia/Shanghai).

Next

Failures · Scheduled runs · Team playbook · Plan → act → check · Human-in-the-loop checkpoints

Pair with today: open the first Next link (agents-failure-modes-and-retries) only after you dated a refuse-line for this guide (Asia/Shanghai).

← Back to home · Learn · Tools · Compare · Start