| name | agent-loop-task-progress |
| description | Record useful working memory during an agent-loop task, including decisions, checkpoints, validation results, scope changes, and blocked states. |
Agent Loop Task Progress
Use this skill while implementing a task after it has been started with
agent-loop-task-start and before it is closed with agent-loop-review-close.
Fast Path
Record important decisions:
vendor/bin/agent-loop session record <task-id> \
--kind decision \
--title "Keep change scoped" \
--body "Only update the dispatcher routing; do not change recall compiler behavior."
Record progress checkpoints:
vendor/bin/agent-loop session checkpoint <task-id> \
--title "Validation" \
--body "vendor/bin/agent-loop init validate --kind=skills passed."
Inspect current task memory:
vendor/bin/agent-loop session show <task-id>
vendor/bin/agent-loop workflow status <task-id>
What To Record
Record:
- decisions that affect implementation direction
- assumptions that future agents must not rediscover
- validation results
- scope changes
- blockers and their cause
- explicit tradeoffs
- commands that materially changed confidence
- why a risky shortcut was not taken
- why a risky shortcut was taken, if it later becomes accepted risk
Do not record:
- raw unbounded logs
- giant diffs
- copied stack traces unless the exact line matters
- noisy command output
- vague notes like "fixed stuff"
- private secrets, tokens, cookies, credentials, or production data
Record Kinds
Use --kind decision when the note changes or explains the direction of the work.
Use checkpoints for:
- validation
- implementation milestones
- review-readiness
- blocked state
- handoff to another agent or human
If the exact supported --kind values differ in a host repo version, prefer
the existing command help and keep the note type conservative.
When To Checkpoint
Checkpoint:
- after selecting the implementation approach
- after touching risky code
- after each meaningful validation run
- before switching files or scope
- before review/close
- when blocked
Scope Changes
If the task changes scope, record it immediately:
vendor/bin/agent-loop session checkpoint <task-id> \
--title "Scope change" \
--body "The task expanded from docs-only to docs plus init validate examples because the existing README list was stale."
Then run:
vendor/bin/agent-loop workflow status <task-id>
If the scope changed enough that recall context is stale, use
agent-loop-l2-context and recompile recall.
Validation Notes
Good validation checkpoint:
vendor/bin/agent-loop session checkpoint <task-id> \
--title "Validation" \
--body "vendor/bin/phpunit --filter Init passed; vendor/bin/agent-loop init validate --kind=all passed."
Bad validation checkpoint:
tests ok
Be specific enough that the next agent knows what was actually run.
Structured Completion Evidence
A validation checkpoint explains progress but does not satisfy a governed
done close. After each required command in the current work-brief revision
has actually run, record the result separately:
vendor/bin/agent-loop session validation record <task-id> \
--brief-revision <current-revision> \
--command "vendor/bin/phpunit tests/FocusedTest.php" \
--status passed --exit-code 0 --by <actor>
Use the exact command string from the brief. Add --duration-ms only when it
was measured. A re-plan creates a new revision, so prior evidence remains
auditable but cannot satisfy the revised validation requirement.
Noise Control
Keep session memory compact. If command output is large, summarize the finding
and point to the command instead of pasting the full output.
Prefer rg for repository search after verifying ripgrep is installed:
rg --version
Prefer RTK-wrapped commands in host repos when output is noisy:
rtk test vendor/bin/phpunit --filter Init
rtk err vendor/bin/phpstan analyse --memory-limit=1G
Do not assume RTK compresses nested Make or Docker output unless the host repo
documents that boundary.
Before Review And Close
Before using agent-loop-review-close, record a final checkpoint:
vendor/bin/agent-loop session checkpoint <task-id> \
--title "Ready for review" \
--body "Implementation complete; recall was recompiled after docs changes; validation passed."
Then continue with:
vendor/bin/agent-loop review blindspots <task-id>
vendor/bin/agent-loop verify
vendor/bin/agent-loop workflow status <task-id>
Record the explicit session learning decision before the governed close:
vendor/bin/agent-loop session learning decide <task-id> \
--status no_durable_learning --by <actor> \
--reason "No reusable finding from this bounded task."
This records an outcome; it does not create or approve durable guidance.
Skill Boundary
This skill owns:
- useful session records during implementation
- compact checkpoints
- validation notes
- scope-change notes
- blocked-state notes
- handoff notes
This skill does not own:
- starting the task (see
agent-loop-task-start)
- compiling L2 context (see
agent-loop-l2-context)
- review/verify/close (see
agent-loop-review-close)
- durable learning promotion
- developing
agent-loop itself
Validation
vendor/bin/agent-loop session show <task-id> shows useful notes
vendor/bin/agent-loop workflow status <task-id> still resolves the task
- final checkpoint exists before review/close
- no secrets or giant logs were stored
Example Triggers
- "Record this decision in the agent-loop session."
- "Checkpoint the validation result."
- "We changed scope; update the task memory."
- "Show the current session notes."
- "Prepare this task for review."