v0.7.0

@jfmoe/smart-search 0.7.0

Provider Credential Pool

• Adds multi-credential round-robin for allowlisted providers: Exa, Tavily, Jina, Firecrawl, Context7, and AnySearch.
• Supports *_API_KEYS JSON arrays (config and env). A non-empty KEYS list fully replaces the matching single *_API_KEY after empty-strip and dedupe.
• Selects credentials with a config-dir claim-advance cursor under file lock, and rotates within the same call only on rate_limited / explicit quota exhaustion.
• Reports pool enablement, key count, and masked tails in doctor/config output without raw secrets.
• Treats KEYS-only pools as configured for capability gates, including AnySearch automatic Vertical Discovery and Jina’s standard minimum-profile contribution.
• Setup and config set manage whole KEYS arrays; no dedicated add/remove key subcommands.

Domain and docs

• Records glossary terms Provider Credential Pool vs Cross-account Quota Aggregation and ADR-0005.
• Skill and README language describes multi-credential load spreading and rate-limit rotation without free-account harvesting framing.

Validation

• Full Python suite: 500 passed, 4 opt-in tests skipped.
• Mock smoke: ok.
• Skill mirror and release metadata checks pass.

v0.6.0

@jfmoe/smart-search 0.6.0

Caller-owned intent routing

• Adds complete Caller Capability Declarations to ordinary search and diagnostic route commands, including catalog-ordered CSV values and the explicit none sentinel.
• Treats a supplied declaration as authoritative: local rules, URL recognition, strict validation, embeddings, and the classifier do not add capabilities. Calls without a declaration retain the existing configured intent router.
• Keeps provider selection, same-capability fallback, unconditional main search, parallel extra sources, and the live Deep Research evidence workflow unchanged.

Shared capability contract

• Establishes the runtime Intent Routing Catalog as the single source for capability identity, semantics, rule terms, semantic examples, classifier material, and generated Skill reference content.
• Updates the Agent Skill and offline Deep Planner so ordinary searches declare the complete capability set in their first search call without a route preflight.
• Synchronizes public and packaged Skill assets and records the authoritative declaration decision in ADR-0004.

Validation

• Full Python suite: 468 passed, 4 opt-in tests skipped.
• Source regression: 290 passed.
• Mock smoke: 27 cases, 0 failed, 0 degraded.
• Skill mirror and generated capability reference checks pass.
• Independent Codex CLI E2E confirmed Documentation Search uses only docs_search and does not attempt Vertical Search or AnySearch.

v0.5.1

@jfmoe/smart-search 0.5.1

Packaging fix

• Includes the versioned AnySearch Verified Domain Manifest in the npm package,
  restoring doctor and search startup for globally installed CLIs.
• Extends the npm package-content gate to require the manifest, preventing future
  releases from passing validation with this runtime asset missing.

Validation

• Reproduced the 0.5.0 installed-package failure and verified the corrected
  package contains and loads the manifest.
• npm test: 442 passed, 4 opt-in live acceptance tests skipped.
• Release metadata, public/package Skill mirror, wrapper, and npm package-content
  checks pass.

v0.5.0

@jfmoe/smart-search 0.5.0

Search Result Journal

• Adds an opt-in, default-disabled Search Result Journal for every completed
  Default Search Invocation through search or s, including successful,
  structured-failure, and CLI-timeout results.
• Writes one complete normalized terminal result per compact UTF-8 JSONL record,
  preserving primary and extra sources, URLs, routing decisions, provider
  attempts, fallback metadata, counts, timing, and failure details.
• Sanitizes a copy before persistence by redacting credential-bearing fields and
  configured credential values without changing stdout or explicit output files.

Local safety and operations

• Uses daily files under the resolved log directory, user-only permissions where
  supported, a 0.5-second cross-process lock, and synchronous non-fsync appends.
• Defaults to 30-day retention, supports permanent retention with 0, and only
  cleans up regular files in the recognized dated Journal namespace.
• Keeps Journal failures non-fatal: one stderr warning, no retry or partial line,
  and no change to the original search result or exit code.
• Extends JSON and human-readable doctor output with enablement, retention,
  resolved directory, and writable/readiness diagnostics.

Validation

• npm test: 442 passed, 4 opt-in live acceptance tests skipped.
• Focused CLI and storage-contract coverage for formats, aliases, output files,
  timeouts, redaction, failure isolation, retention, permissions, and concurrent
  writers.
• Public and packaged Skill mirror verification, release metadata checks, and npm
  package-content dry-run.

v0.4.0

@jfmoe/smart-search 0.4.0

AnySearch Acceptance Surface

• Aligns explicit AnySearch operations with the current MCP tools for Domain
  Discovery, domain-less Vertical Discovery, explicit domain search, Batch
  Discovery, and AnySearch Extraction.
• Supports nested JSON Sub-domain Parameters while keeping operation names,
  upstream tool names, schema-validation state, and classified errors observable.
• Returns compact search candidates with title, URL, and descriptions capped at
  300 characters; full page content remains an explicit fetch step.

Vertical Discovery and domain status

• Keeps automatic AnySearch discovery behind vertical intent, provider filters,
  balanced or strict validation, and an explicit ANYSEARCH_API_KEY.
• Adds focused academic, gaming, and travel intent coverage without implementing
  Automatic Domain Search or changing Web Search fallback behavior.
• Introduces an initially empty Verified Domain Manifest and truthful per-operation
  status. Candidate domains remain discovered/unverified until all acceptance
  gates are satisfied.

Current explicit domain contracts

• Refreshes the observed request parameters for security.vuln,
  finance.fundamental, and code.doc after live MCP contract drift.
• Keeps AnySearch optional and experimental; it does not satisfy the standard
  minimum profile or become Web Search or Web Fetch.

Validation

• npm test: 430 passed, 4 opt-in live acceptance tests skipped.
• Public and packaged Skill mirror verification.
• npm package-content dry-run and release metadata checks.
• Live E2E comparison across academic, security, finance, code, gaming, and travel
  scenarios, with compact AnySearch results and selective fetch verification.

v0.3.0

@jfmoe/smart-search 0.3.0

Persistent Skill management

• Replaces one-shot tool targets with a saved Skill Installation Preference.
• Adds skills install, status, update, and clear for the Agents, Claude,
  Hermes, and custom Skill Containers.
• Normalizes and validates saved paths while preserving user-added and obsolete
  files during managed updates.

Automatic Skill Sync

• Synchronizes the bundled smart-search-cli Skill on the first ordinary CLI
  invocation after an exact version-string change.
• Keeps successful maintenance silent and confines recoverable failure guidance
  to standard error without changing the requested command result.
• Serializes concurrent processes and atomically replaces configuration and
  managed Skill files.
• Lets interactive setup establish the same preference while non-interactive
  setup and --skip-skills preserve existing preferences.
• Keeps package installation separate from writes to user Skill Containers.

Validation

• npm test
• Public CLI and real cross-process synchronization acceptance tests.
• Standards and specification review with confirmed findings repaired.
• Public and packaged Skill mirror verification.
• Mock smoke and npm package-content dry-run checks.

v0.2.3

@jfmoe/smart-search 0.2.3

Context7 result guidance

• Documents that context7:/... items returned by ordinary search are
  library candidates, not fetchable or citable sources.
• Directs agents to take the first relevant candidate, strip the context7:
  prefix, and retrieve documentation with context7-docs.
• Clarifies that a successful Context7 docs_search attempt means library
  resolution succeeded; it does not mean documentation content was retrieved.
• Preserves the existing search result schema and runtime behavior.

Validation

• python3 scripts/sync-skill.py --check
• smart-search regression
• smart-search smoke --mock --format json
• npm test
• Live Codex CLI validation with gpt-5.6-luna confirmed the first Context7
  candidate was passed to context7-docs instead of being fetched or cited.

v0.2.2

@jfmoe/smart-search 0.2.2

Search timeout

• Raises the default smart-search search hard timeout from 90 to 180 seconds.
• Updates the public Skill, packaged Skill, and bilingual documentation to use the
  180-second search budget consistently.
• Adds regression coverage for the CLI default and distributed Skill contract.

Validation

• npm test
• smart-search regression
• smart-search smoke --mock --format json
• Release metadata verified consistently across npm, lockfile, and Python
  package metadata.

v0.2.1

@jfmoe/smart-search 0.2.1

Skill discovery

• Publishes the updated Skill contract as @jfmoe/smart-search@0.2.1.
• Advertises X/Twitter search in the public Smart Search Skill description and
  research-or-retrieval task branch.
• Keeps the packaged Skill mirror synchronized with the public source.

Validation

• npm test
• Release metadata verified consistently across npm, lockfile, and Python
  package metadata.

v0.2.0

@jfmoe/smart-search 0.2.0

Personal release identity

• Moves the user-facing npm package to @jfmoe/smart-search@0.2.0.
• Aligns npm metadata, Python metadata, lock metadata, repository links, install
  and repair guidance, README documentation, and the public Smart Search Skill.
• Formally supports macOS with Node as the only promised JavaScript launcher.

Release policy

• A normal main push only runs tests.
• Preview publishing is a manual workflow_dispatch action pinned to a full
  40-character commit SHA and publishes only to next.
• Stable publishing requires an exact v0.2.0 tag and verifies version
  consistency before publishing to latest.

Validation

• npm test
• npm pack --dry-run through the package-content check in npm test
• No release, tag, registry, or workflow-dispatch action was performed while
  preparing these notes.