cookbook/filesystem/_05_operations/TEST_LOG.md
Tested 2026-07-24, agno 2.8.1 (source tree, branch feat/agent-fs at 7df2fad3a). No model and no API keys, since both files are pure Python against the store.
Entries quote printed state and tool calls. Both files are pure Python against the store, so their output is deterministic.
Re-run 2026-07-24 after the FileSystem(db) change and the quota-message rewording.
Entries quote printed state verbatim; both files are deterministic.
Status: PASS
Description: Hits the per-file cap and the namespace cap on purpose (caps shrunk to 200/300 bytes), prints the typed errors and the exact tool strings an agent would see, then recovers by partitioning and by deleting the oldest partition.
Result: Per-file cap gave typed error file 209 > 200 and tool string "Error: seen/2026-07-22.md would be 209 bytes (limit 200 per file). Start a new file (for record logs, partition by date, e.g. seen/2026-07-24.md) or delete files you no longer need." Recovery 1 wrote the record to a new partition. Namespace cap gave typed error namespace 284 of 300 and tool string "Error: storage is full (284 of 300 bytes). Delete only files you are certain are obsolete (see list_files), such as an old date partition, then retry. Do not overwrite or delete records you might still need to make room; if nothing is safely disposable, stop and report that storage is full." Recovery 2 deleted the oldest partition (3 files/284 bytes -> 2 files/133 bytes) and the retried append succeeded.
Status: PARTIAL
Description: Drives a file store from a plain script: lists files with sizes and versions, reads working state, seeds seen-records, and verifies them with contains().
Result: Every operation works and the output is deterministic across runs. Listing showed seen/2026-07-23.md 39B v1 and state/last-run.md 30B v1, usage {'files': 2, 'bytes': 69}. Read back "2026-07-23: briefed 3 stories". After seeding two more ids, the membership check returned {'found': ['kite-os-release'], 'missing': ['brand-new-story']}.
Gap: the example does not demonstrate its headline claim. It mints a throwaway tmp/agent_fs_inspect_<uuid>.db, seeds that store itself, and then inspects the data it just wrote, all in one process. Nothing here shows a script reaching a store some agent wrote, so every printed line would be identical if FileSystem had no cross-process sharing at all. The capability is real and covered by libs/agno/tests/; this file just does not put it on screen. Pointing it at tmp/filesystem/getting_started.db, the one store an agent actually writes in this cookbook, would close the gap.