Chronicle

Chronicle separates observed events from the intent only the person or agent doing the work can state. Capture requires installed, active hooks; check coverage before assuming the trace is complete.

Resume first

Run chron resume in the project root. Read unresolved questions and state declarations. Then record what you are taking up:

chron open "continue search" --state "empty results are intentional staging"

If the CLI is missing, install with pipx install 'git+https://github.com/AntreasAntoniou/chronicle.git'. Installing the skill does not install hooks or establish capture coverage.

Five moments worth recording

Command	What to record
chron open	Your belief about current state and what you are taking up
chron arm	The operation, intent, reversibility class, and verified restore path
chron decision	The choice, reason, and any ambiguous state that is intentional
chron landed	A verified external change with resource/version and rollback references
chron close	Where work stopped, what remains undone, and open questions

Before a destructive or bulk operation:

chron arm "replace the staging index" \
  --intent "load the verified replacement" \
  --class R1 \
  --restore "snapshot path and tested restoration command" \
  --verified "restoration tested on a disposable copy"

R0 is fully reversible, R1 needs a recovery artifact, and R2 is irreversible. An ARM does not grant user authority or prove that a host gate is active. Follow the user's scope and verify hook integration separately.

After publication or deployment, record observed state using chron landed --ext and --verified. A successful command alone does not prove an external change.

Correct and recover

Append chron correct with the original entry ID, truth, and evidence. Preserve history.

chron history src/app.py
chron show src/app.py --at 2h
chron restore src/app.py --at 2h --to /tmp/recovered-app.py
chron search "search index"
chron doctor

Only captured versions can be recovered. State gaps explicitly. Use chron experiment and chron abandoned for attempts and dead ends, quoting measured results only.

Inference and privacy

• Narrated entries marked inferred are hypotheses. Inspect anchored events before acting; append a firsthand statement when you can resolve an uncertainty.
• Narration requires an explicit model, can incur cost, and sends trace content to that provider. Review its dry-run prompt first.
• Keep secrets out of narrative entries and keep the ledger, CAS, transcripts, and canvas private. Redaction does not guarantee detection of every embedded secret.
• Keep operational history here; route commitments to a task tracker and lasting knowledge to canonical project documentation.

Read references/codex.md before experimenting with Codex hooks. Configuration, trust, execution, and verified coverage are separate states.