TestIsContextLengthError verifies detection of context-length errors from real API responses. Each test case is based on actual error strings from Ope

Verify that isObviousPrivateHost catches common private hostnames.

Verify that public hostnames pass the pre-flight check.

--------------------------------------------------------------------------- Budget check tests -------------------------------------------------------

TestIsOverBudget_OverBudget verifies that a large history exceeding a small context window returns true. We use a tiny window (500 tokens) and fill hi

TestIsOverBudget_SummaryCountedInBudget verifies that the summary field is included in the budget calculation. A large summary should push the total o

TestIsOverBudget_UnderBudget verifies that a small history within a generous context window returns false.

Verify that IPv4-mapped IPv6 loopback is blocked.

Verify that common private IPv4 ranges are blocked.

Verify that public IPv4 addresses are allowed.

Verify that IPv6 private/restricted addresses are blocked.

Verify that public IPv6 addresses are allowed.

Verify that nil IP is treated as restricted.

ListUserWorkspaces decodes the JSON array correctly and preserves order.

Malformed JSON body is a 400 before we ever talk to the service.

Wrong credentials produce ErrInvalidCredentials, not ErrUnauthorized, so the CLI can distinguish "wrong password" from "stale token".

Wrong credentials surface as 401 with a generic message; no token is issued and the response says nothing about which field was wrong.

Missing email or password is a 400 with a distinct message.

Secret loader failure after a successful password check is a 500: the client did everything right, the server cannot finish the job.

--- tests --- Happy path: Login returns a token, stores it internally, and decodes the user record correctly.

Happy path: valid credentials produce a token that parses back to the same user, plus a JSON body with token + user.

Verify that looksLikeHTML correctly detects HTML content.

TestMain lets the test binary impersonate the astroclaw CLI when testscript exec's "astroclaw <args>" from a .txtar script. When run normally, it fall

Verifies that bad JSON arguments return an error.

Verifies that empty content returns an error.

Verifies that store errors propagate to the caller.

Verifies that a valid memory is saved and the tool returns success.

OPENAI_API_KEY=sk-xxx go test ./pkg/provider/ -v -run TestOpenAIChatStream

OPENAI_API_KEY=sk-xxx go test ./pkg/provider/ -v -run TestOpenAIChatSync

Test HTML parsing with a snapshot of DuckDuckGo's HTML structure. If DuckDuckGo changes their HTML, update this snapshot and the regexes.

Verify that the count parameter limits how many results are returned.

Verify that HTML with no matching result elements returns an empty slice.

Expired token: Parse must surface ErrTokenExpired specifically, so the middleware can distinguish "log in again" from "this token was always bad".

alg=none attack: an attacker crafts a token with no signature. Parse must refuse it; otherwise anyone could forge claims. This is the canonical JWT vu

Tampered payload: flip a byte in the middle base64 segment. The signature won't match the modified payload and Parse must reject.

Wrong secret: Parse with a different secret must fail because the HMAC signature won't match. Catches "someone reused a key they shouldn't have".

TestParseTurnBoundaries_Empty verifies that an empty history produces no boundaries.

--------------------------------------------------------------------------- Turn boundary tests ------------------------------------------------------

TestParseTurnBoundaries_SingleUser verifies a history with just one user message produces exactly one boundary.

TestParseTurnBoundaries_WithToolCalls verifies that tool_call and tool_result messages do NOT create new boundaries. They belong to the same Turn as t

Expired tokens are rejected even though the signature is valid.

Protected routes require a bearer token. Without one, the auth middleware should reject the request with 401 before any service is called.

TestPushRoundup proves the full round trip: client upgrades against the Handler, Registry.GetByUser (real DB row) surfaces the connID, Bus.Send writes

TestReadFile_BadJSON verifies that malformed JSON arguments return an error rather than panicking.

TestReadFile_BinaryFile verifies that files containing null bytes are rejected with an error. Binary files (executables, images, etc.) would produce g

TestReadFile_EmptyFile verifies that an empty file returns an empty string without error. This is a valid case — the model should see "" and can tell

Verify that individual lines exceeding 2000 characters are truncated with a suffix, while the rest of the file is returned normally.

TestReadFile_NormalText verifies the happy path: a small text file is read in full, with contents returned exactly as written.

TestReadFile_NotFound verifies that a non-existent file returns an error containing the file path, so the model can tell the user which file was not f

TestReadFile_Truncation verifies that files exceeding 2000 lines are truncated and the result ends with a [TRUNCATED] marker. This prevents a single l

Verify that a file with few lines but exceeding 50KB is truncated at the byte limit, not the line limit.

Short secrets are refused by both Sign and Parse so a misconfigured deployment can't silently produce weak HMACs.

TestReply_InvalidSession verifies that Reply with a non-existent session returns an error.

TestReply_PersistsMessages verifies that Reply saves both the user message and assistant reply to the database.

TestReply_RestoresContext verifies that context is correctly restored when creating a new Agent from a session with existing ContextMessages.

TestReply_SavesContext verifies that after Reply, the session's ContextMessages are updated, so the next Reply can restore Agent state.

Expired token: 401 with "token expired" so the client can distinguish "log in again" from "this token was always bad".

Tampered payload: signature won't verify, falls into the generic invalid bucket.

No Authorization header at all: 401 and next must not run.

Secret loader failure: 500 (server fault, not client fault) and next must not run.

Happy path: a valid token passes through and the handler sees the claims.

Non-Bearer scheme is rejected with the same 401.

Verifies that runMigrations() produces the same schema as executing migration SQL files directly. Compares columns (name, type, nullable, default), in

Double check: run runMigrations() again on an already-migrated database. Statements should be skipped via the tracking table, or if tracking fails, ca

Verifies that the _migrations tracking table has the correct number of records (one per statement in the migration files).

TestScripts runs every .txtar in testdata/scripts/ against a real deployed backend. URLs and admin password come from ASTROCLAW_API_URL, ASTROCLAW_REP

TestSendMissingConn checks the ErrConnGone signal that the reply pipeline uses to unregister stale connections. This one needs no Registry — Bus alone

Round-trip: a token produced by Sign must Parse back into the same logical claims. Times are compared with truncation to one second since JWT encodes

Verifies that comments preceding a SQL statement are preserved as part of the statement, not stripped or treated as a separate block.

Verifies that comment-only blocks (no actual SQL) are filtered out. Atlas may prepend directives like "-- atlas:txmode none" that should not be treate

Verifies that an empty input returns an empty slice, not nil or panic.

--- Unit tests for splitStatements --- Verifies that multiple DDL statements separated by semicolons are correctly split into individual statements.

Verifies that a single statement without any extras is returned as-is.

Verifies that a trailing semicolon at the end of the file does not produce an extra empty statement.

Verify that stripHTML removes tags and normalizes whitespace.

TestSummarizeOldTurns_AppendsToExistingSummary verifies that new summaries are appended to existing ones, preserving earlier context.

--------------------------------------------------------------------------- LLM summarization tests --------------------------------------------------

TestSummarizeOldTurns_IncludesToolResults verifies that small tool results are included in the summarization input, not filtered out.

TestSummarizeOldTurns_LLMFailure verifies that when the LLM fails, the fallback summary is used instead. History is still truncated.

TestSummarizeOldTurns_TooFewMessages verifies that summarization is skipped when history has 4 or fewer messages.

Verify that truncateBytes respects UTF-8 character boundaries when truncating at the byte limit.

Verify that truncateBytes returns the original string when under the limit.

TestWriteFile_BadJSON verifies that malformed JSON arguments return an error rather than panicking.