Skip to content

feat!: Consolidate Taskless skills into one router + recipe-driven CLI help - #18

Merged
thecodedrift merged 20 commits into
mainfrom
jakob/tskl-231
May 12, 2026
Merged

feat!: Consolidate Taskless skills into one router + recipe-driven CLI help#18
thecodedrift merged 20 commits into
mainfrom
jakob/tskl-231

Conversation

@thecodedrift

@thecodedrift thecodedrift commented May 11, 2026 •
edited
Loading

Copy link
Copy Markdown
Member

Collapses the 10 per-task Taskless skills + 6 slash commands into a single taskless skill plus a single /tskl router. Per-task instructions move out of always-loaded skill bodies and into npx @taskless/cli help <topic> recipes, fetched on demand.

A customer reported that loading the Taskless plugin caused the Claude Code harness to evict other skills from the working set. Codex behaves similarly. The 1,367 lines previously committed across 10 SKILL.md files + 6 near-duplicate slash command files were inflating the agent's always-loaded surface even when the user wasn't doing anything Taskless-related. Worse, the per-skill triggers were inconsistent — taskless-create-rule listed phrases like "create a rule", "add a lint rule", "detect this pattern" with no Taskless anchor, so it over-fired in any repo that uses ESLint.

This change is the result of a long design conversation captured in openspec/changes/archive/2026-05-10-consolidate-taskless-skills/ (proposal, design, tasks, 16 capability spec deltas). The high-bit decisions:

Skill body becomes a router, not a recipe. The new skills/taskless/SKILL.md is ~50 lines: it tells the agent it does NOT have the steps, instructs it to fetch the canonical recipe via npx @taskless/cli help <topic>, and lists the available topics. Recipes live in the CLI bundle, so agents always read the version current to the installed CLI rather than whatever was frozen into a SKILL.md at install time.

Trigger description anchors on Taskless. The single description requires the user to reference Taskless explicitly OR for .taskless/ to be referenced. Generic ESLint/lint/rule phrasing without that anchor SHALL NOT trigger. The skill body's first step is a .taskless/ presence check with graceful failure ("ask the user to confirm they meant Taskless"). This is strictly better than today's mixed-bag where rule-create/improve/delete fire promiscuously.

--anonymous is global, with a per-command behavior matrix (option A from the design discussion). On info it skips the API/auth probe; on auth login it errors with "auth commands cannot be anonymous"; on rule create/rule improve the CLI exits with a pointer to the local-only recipe variant (the agent owns the local generation flow per the recipe — we did NOT bundle Claude SDK in the CLI). Everywhere else it's accepted as a no-op so agents don't have to remember which commands accept it.

Anonymous variants live as <topic>.anonymous.txt alongside the canonical recipes. Build-time map keeps lookup O(1); fetching taskless help rule create --anonymous returns the variant when one exists and falls back to the canonical recipe otherwise. No central registry to keep in sync.

Recipe template is fixed. Every tskl help <topic> returns the same shape: # Topic: <name> (CLI v<x.y.z> / topic v<n>) header, ## Goal, ## Preconditions, ## Steps, optional ## Input schema (JSON Schema rendered from Zod via z.toJSONSchema() and interpolated through a {{INPUT_SCHEMA}} placeholder), ## Errors (mapping stable error codes to user-facing fixes), ## See Also. The fixed shape lets agents pattern-match consistently.

Standardized JSON error envelope. When --json is set, failures emit { ok: false, code: \"<CODE>\", message: \"<...>\" } with stable codes (AUTH_REQUIRED, NO_GITHUB_REMOTE, RULE_GENERATION_FAILED, RULE_NOT_FOUND, INVALID_INPUT, NETWORK_ERROR, SCAN_FAILED, INTERNAL_ERROR). Recipes reference these codes by name in their ## Errors sections so agents can branch on the code field rather than parsing English error strings.

CLI verb rename: rules → rule (singular) so tskl help rule create reads naturally and the next CLI command the agent runs is the same noun (taskless rule create). No compatibility alias — the plural form errors cleanly.

Telemetry hard-renamed to help_<topic> (intent), help_index (no-args fetch), help_unknown (probable confusion), cli_<action> (action started), cli_<action>_completed (success + durationMs). The funnel makes wrong-topic re-routing observable as a derivable signal: a help_<topic_a> event followed by no cli_<action_a> and a subsequent help_<topic_b> indicates the agent fetched, didn't act, and re-routed.

taskless update as its own subcommand. Same logic as init --no-interactive but exposed as a dedicated verb so agents can refresh installs without explaining flags.

Migration is automatic. The existing state.ts install plumbing records what was written per target. On next npx @taskless/cli, the install reads previous state, computes the diff (10 obsolete skills + 6 obsolete commands removed, 1 new skill + 1 new command added), confirms with the user, applies, and reports the cleanup. Integration test in apply-install-plan.test.ts proves the v0.6 → v0.7 path end-to-end.

Out of scope for this change (deliberate non-goals from the design):

  • Cursor/Codex formal skill+command support (deferred — known to evict similarly, so the consolidation already helps them; formal support comes later)
  • Bundling Claude SDK in the CLI for fully-self-contained anonymous generation (Option B from the design discussion — rejected; the LLM is the generator, moving that into the CLI is a much larger change)
  • Soft deprecation of v0.6 skill names (hard cut at v0.7)
  • JSON-only tskl help output (markdown is sufficient for agents)

The full motivation, alternatives considered, decisions, risks, and open questions are captured in openspec/changes/archive/2026-05-10-consolidate-taskless-skills/{proposal,design}.md.

Closes TSKL-231 (internal feedback)

thecodedrift and others added 15 commits May 10, 2026 21:46
@thecodedrift @claude
docs(openspec): Add consolidate-taskless-skills change proposal
3a554b2 
Adds the OpenSpec change artifacts (proposal, design, tasks, spec
deltas) for collapsing the 10 per-task Taskless skills into a single
consolidated `taskless` skill plus single `tskl` command. Drives
recipes through `npx @taskless/cli help <topic>` to remove always-loaded
context bloat.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
fix(cli): Remove --schema flag and printSchema utility
77d17b2 
Removes the user-facing --schema flag from every CLI command (info,
check, rules create/improve/meta/verify). Schemas will be embedded
inline in `tskl help <topic>` output via zod-to-json-schema in a
follow-up. Zod schemas themselves remain as the source of truth.

Deletes packages/cli/src/util/schema-output.ts and the schema.test.ts
integration tests. Drops the verify --schema CLI integration test;
the underlying getSchemaPayload() function and its unit tests stay
since the recipe will reuse the payload content.

Part of consolidate-taskless-skills (task group 12).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
feat(cli): Rename telemetry events to help_*/cli_*_completed taxonomy
6892f40 
Renames the help command's events to surface agent intent:
  cli_help        → help_index   (agent fetched the topic list)
  cli_help_<topic>→ help_<topic> (agent fetched a specific recipe)
  (new)           → help_unknown (agent asked for a missing topic)

Adds cli_<action>_completed events to every action command (info, check,
rule create/improve/delete/meta/verify, auth login/logout/status) with
success and durationMs properties. Hard rename — no dual-emit window.

Wraps each command's run() in try/finally so the completion event fires
regardless of whether the command returned normally or threw via the
local fail() helper.

Part of consolidate-taskless-skills (task group 11).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
feat(cli)!: Rename rules subcommand to rule (singular)
87d6a6c 
BREAKING CHANGE: `taskless rules <command>` is no longer recognized.
Users must invoke `taskless rule create`, `taskless rule improve`,
`taskless rule delete`, `taskless rule verify`, `taskless rule meta`.

Aligns the CLI verb with how recipes will address commands
(`tskl help rule create` reads naturally only if the CLI takes the
singular form). The internal source file `commands/rules.ts` keeps
its name; the citty subcommand registration switches to `rule`.

Renames help files (rules-*.txt → rule-*.txt) and updates user-facing
references in error messages, README, and help text. Internal
filesystem paths like `.taskless/rules/` stay (that's a directory of
rules, not the verb).

Part of consolidate-taskless-skills (task group 2).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
feat(cli): Standardize error envelope with stable error codes
9e127cc 
Adds packages/cli/src/types/errors.ts defining the stable CliErrorCode
union (AUTH_REQUIRED, NO_GITHUB_REMOTE, RULE_GENERATION_FAILED,
RULE_NOT_FOUND, INVALID_INPUT, NETWORK_ERROR, SCAN_FAILED, INTERNAL_ERROR)
and a makeErrorEnvelope() helper.

When any action command exits with an error and --json is set, the
output is now the standardized envelope:
  { "ok": false, "code": "<CODE>", "message": "<...>" }

Migrates fail() helpers in rules create/improve/meta/verify and the
catch path in check to emit codes:
  - INVALID_INPUT for --from validation, missing rule ID, bad metadata
  - AUTH_REQUIRED / NO_GITHUB_REMOTE for identity resolution
  - NETWORK_ERROR for API submit/poll
  - RULE_GENERATION_FAILED for terminal API failures
  - RULE_NOT_FOUND when rule metadata is missing
  - SCAN_FAILED for ast-grep scan errors

Recipes will reference these codes by name in their `## Errors`
sections so agents can branch on them.

Part of consolidate-taskless-skills (task group 6, code paths complete;
unit tests for code-emission deferred to task 6.4 follow-up).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
feat(cli): Add global --anonymous flag with per-command behavior
52dff7e 
Recognized on every command. Behavior matrix:
  rule create / rule improve  → exit with pointer to local-only recipe
                                (per Option A: generation runs in the
                                agent via `taskless help <topic> --anonymous`,
                                not in the CLI itself)
  rule delete / verify / meta → accepted as no-op (purely local)
  check                       → accepted as no-op (no auth dependency)
  info                        → skip the API/auth probe, report local
                                state only
  auth login                  → exit 1 with "auth commands cannot be
                                anonymous"
  auth logout                 → accepted as no-op (logout is local)
  init                        → accepted as no-op

The flag is universally accepted by the parser so agents never have to
remember which commands accept it. Tests for the per-command matrix
deferred to task 3.7 follow-up.

Part of consolidate-taskless-skills (task group 3).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
feat(cli): Simplify wizard and route non-TTY bare invocation to help
63040f0 
Removes the optional-skills wizard step (`steps/optional-skills.ts`)
and the corresponding `promptOptionalSkills()` call in `runWizard()`.
With v0.7's catalog reducing to a single mandatory skill, there's
nothing to opt into.

Bare `taskless` without a TTY now prints a short context preamble
followed by the topic index from `help`, so agents and pipes get
something useful instead of citty's default usage screen.

Init reporting now lists obsolete skills and commands removed during
re-install (the underlying state.ts already tracked them; this just
surfaces the removals so users see the cleanup).

Updates wizard-integration tests to drop optional-skills mocks and
restructure the summary-cancel test to seed the manifest with a stale
skill name (instead of relying on the optional-skill diff).

Part of consolidate-taskless-skills (task group 10).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
feat(cli): Extend help command with variants, schema embedding, and slug
10ad745 
Adds three capabilities to `taskless help`:

1. Anonymous variant lookup. Filename convention <topic>.anonymous.txt
   marks a local-only recipe. When `--anonymous` is passed, the help
   command serves the variant if present and falls back to the
   canonical recipe otherwise. The variant set is derived at build
   time via import.meta.glob (O(1) lookup, no central registry).

2. Recipe placeholders. {{CLI_VERSION}} interpolates the build-time
   CLI version into the recipe header. {{INPUT_SCHEMA}} interpolates
   a JSON Schema rendered from the topic's Zod input source via
   z.toJSONSchema() (zod 4 built-in — no zod-to-json-schema dep).
   Currently wired for rule-create and rule-improve.

3. No-args index now leads with a human slug ("for agents… for
   humans…") and ends with the --anonymous note. Same machine output
   shape, friendlier preamble for developers who wander into it.

Telemetry: topic events now include `anonymous: boolean` so the
funnel can distinguish authed vs local-only intent.

Recipe content authoring is deferred to task 8.

Part of consolidate-taskless-skills (task group 7).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
feat(skills)!: Consolidate 10 skills + 6 commands into single taskles…
b4da407 
…s skill

BREAKING CHANGE: Replaces the 10 per-task skills (taskless-check,
taskless-ci, taskless-create-rule, taskless-create-rule-anonymous,
taskless-delete-rule, taskless-improve-rule, taskless-improve-rule-
anonymous, taskless-info, taskless-login, taskless-logout) with one
consolidated `taskless` skill, and the 6 slash commands with one
`/tskl` router command.

The new skill is a ~40-line router whose body explicitly tells the
agent it does NOT have step-by-step instructions and must fetch the
canonical recipe via `npx @taskless/cli help <topic>`. The recipes
themselves stay in the CLI bundle (per task 8) so agents always read
the version current to the installed CLI, not whatever was frozen
into a SKILL.md at install time.

The trigger description anchors on Taskless-specific phrases or
.taskless/ directory references and explicitly says "do NOT trigger
on generic ESLint, linting, or rule requests". This tightens the
over-firing of the previous create/improve/delete triggers, which
listed phrases like "create a rule" / "add a lint rule" with no
Taskless anchor.

Catalog reduces to a single mandatory entry (`taskless`).
getOptionalSkillNames() and isOptionalSkill() now return empty / false.
The Vite assertSkillVersions guard already handles arbitrary catalog
sizes.

Existing v0.6 installs auto-migrate via the idempotent reinstall path
in install.ts: the manifest records the 10 old skill names; on next
init, applyInstallPlan reads the previous state, computes removals,
deletes the obsolete files, writes the new single skill + command,
and updates the manifest. Init reporting (added in task 10) surfaces
the cleanup so users see what changed.

Test fixtures (apply-install-plan, wizard-integration, init-no-
interactive, cli) updated to reference the new skill name. The
"does NOT install optional skills" test is removed since there are
no optional skills.

Part of consolidate-taskless-skills (task groups 1 and 9).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
test(cli): Add v0.6 → v0.7 skill consolidation migration integration …
2f9661e 
…test

Adds an integration test in apply-install-plan.test.ts that:

1. Seeds 10 v0.6 skill directories and 6 v0.6 command files on disk
2. Records all of them in the install state manifest (as v0.6 would
   have left it)
3. Runs applyInstallPlan with the v0.7 plan (single taskless skill
   + single tskl.md command)
4. Asserts:
   - 10 obsolete skills + 6 obsolete commands removed from disk
   - new taskless/SKILL.md + commands/tskl/tskl.md present
   - manifest updated to reflect the new layout
   - returned counts match (10 removed skills, 6 removed commands,
     1 written skill, 1 written command)

Proves the idempotent reinstall path correctly migrates v0.6 installs
to v0.7 without requiring a separate migration script.

Part of consolidate-taskless-skills (task group 13).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
docs(cli): Author all topic recipes following the canonical template
f36dea5 
Rewrites every help/*.txt file to follow the recipe template:
  # Topic: <name>     (CLI v{{CLI_VERSION}} / topic v1)
  ## Goal
  ## Preconditions
  ## Steps
  ## Input schema  (where applicable, via {{INPUT_SCHEMA}})
  ## Errors
  ## See Also

Recipes authored:
- rule-create.txt           — API-backed; embeds zod-derived JSON schema
- rule-create.anonymous.txt — local-only; agent-driven verify loop
- rule-improve.txt          — API-backed; embeds iterate input schema
- rule-improve.anonymous.txt — local-only; in-place edit + verify loop
- rule-delete.txt           — local file removal
- rule-verify.txt           — agent-internal, used by anonymous loops
- rule-meta.txt             — agent-internal, used by improve flow
- rule.txt                  — umbrella pointing at the subcommand recipes
- check.txt                 — full + diff scan patterns, JSON output shape
- auth.txt                  — combined login/logout/status with branches
- info.txt                  — local state report
- ci.txt                    — ported from v0.6 taskless-ci skill body
- init.txt                  — points user at `npx @taskless/cli`

Removes the obsolete per-subcommand auth-login.txt / auth-logout.txt
files; auth.txt now covers all three branches.

Each recipe ends with an Errors table mapping the standardized
CliErrorCode values (AUTH_REQUIRED, NO_GITHUB_REMOTE, INVALID_INPUT,
NETWORK_ERROR, RULE_GENERATION_FAILED, RULE_NOT_FOUND, SCAN_FAILED,
INTERNAL_ERROR) to user-facing fixes — agents can branch on the code
field of the JSON error envelope.

Smoke-tested: `taskless help rule create` renders correctly with the
embedded schema; `--anonymous` returns the local-only variant;
`taskless help` prints the new human slug + topic table.

Part of consolidate-taskless-skills (task group 8).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
docs(release): Changeset and README updates for v0.7 consolidation
ae1a133 
Adds a changeset describing the consolidate-taskless-skills change:
- 10 per-task skills + 6 commands → 1 skill + 1 command
- CLI verb rename (rules → rule, breaking)
- --schema flag removed
- Global --anonymous flag added
- Telemetry rename (hard cut)
- Standardized JSON error envelope
- Recipe template + on-demand fetch via `taskless help <topic>`

Migration is automatic on next `npx @taskless/cli` (idempotent
reinstall reads the manifest, deletes obsolete files, writes the new
single skill + command).

Updates the root README to reflect the consolidated structure and the
new "add a topic recipe" workflow (replaces "add a new skill").
Updates packages/cli/README.md with the new init wizard description
(no optional-skills step), `--anonymous` matrix, and `taskless help`
documentation.

Plugin/marketplace JSON version bumps are deferred to `pnpm bump`
(part of the release workflow that consumes the changeset).

Closes the consolidate-taskless-skills OpenSpec change. All 12 task
groups complete; ready for /opsx:archive.

Part of consolidate-taskless-skills (task group 15).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
test(cli): Cover error envelope, --anonymous matrix, and help extensions
7a9ed0f 
Adds three new test files (31 tests total):

error-envelope.test.ts (7 tests):
  Asserts each command's --json failure path emits the standardized
  { ok: false, code, message } envelope with the documented stable
  codes (INVALID_INPUT for missing/bad --from input, RULE_NOT_FOUND
  for missing rule meta, etc.). Also asserts the envelope shape has
  exactly the documented fields.

anonymous-flag.test.ts (10 tests):
  Covers the per-command --anonymous behavior matrix end-to-end:
  - info skips API/auth probe (loggedIn: false)
  - auth login rejects with "auth commands cannot be anonymous"
  - auth logout / check / rule delete / rule verify / rule meta /
    init: accept as no-op (same behavior as without the flag)
  - rule create / rule improve: exit with pointer to local-only recipe

help-extensions.test.ts (12 tests):
  - No-args output: human slug, topic table, --anonymous mention
  - Recipe rendering: {{CLI_VERSION}} and {{INPUT_SCHEMA}} interpolation
  - Anonymous variant: returns .anonymous variant when present, falls
    back to canonical when absent, omitted flag returns canonical
  - Unknown topic exits non-zero
  - Bare taskless (non-TTY) routes to topic index with preamble

Also fixes a small regression in bare-taskless non-TTY: the parent's
`-d <path>` flag was being mis-parsed as a topic name by the help
command. Now forwards the parent's rawArgs explicitly so the help
command's flag-aware positional extraction handles them correctly.

Closes the deferred test items 6.4, 3.7, and 7.7.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
feat(cli): Add taskless update subcommand for non-interactive refresh
10f3697 
Adds a dedicated `taskless update` subcommand that runs the same
non-interactive install path as `taskless init --no-interactive` but
without requiring the agent to know about (or explain) the flag.
Detects installed tools, computes the diff against the recorded
state, writes the consolidated skill, removes obsolete files, and
prints a summary.

Use case: agent or user already has Taskless installed and wants to
refresh to a new bundled version. `npx @taskless/cli update` is the
clean way to ask for that without a wizard popping up.

Emits cli_update / cli_update_completed telemetry events with the
same `locations`, `success`, `durationMs` shape as init.

Updates the consolidated skill body and the /tskl router to include
the new "Update existing install → `npx @taskless/cli update`" row.
Adds `help/update.txt` recipe.

Tests: two new cases in init-no-interactive.test.ts cover the
detected-tools and `.agents/` fallback paths for `update`.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
chore(openspec): Archive consolidate-taskless-skills + sync main specs
c517d48 
Syncs the change's delta specs into the main spec catalog and moves
the change directory to openspec/changes/archive/.

Spec sync summary:

ADDED (new main spec):
  • skill-taskless

MODIFIED (8 capabilities merged):
  • analytics       — telemetry rename to help_*/cli_*_completed
  • cli             — global --anonymous flag + standardized error envelope
  • cli-auth        — --anonymous handling on login/logout + error envelope
  • cli-check       — --anonymous no-op + error envelope
  • cli-help        — anonymous variant lookup, schema embedding,
                      no-args index slug, intent telemetry
  • cli-init        — wizard simplification (no optional skills),
                      install manifest changes, bare-taskless routing
  • cli-rules       — rules → rule rename across every requirement,
                      anonymous local-only branches added
  • skills          — collapsed per-task skills into single taskless
                      skill, plugin manifest, command file conventions

REMOVED (7 capability dirs deleted):
  • cli-flag-schema (--schema flag removed)
  • skill-create-rule, skill-improve-rule, skill-delete-rule,
    skill-auth-login, skill-auth-logout, skill-ci
    (folded into skill-taskless)

Tasks.md cleaned up before archive: tasks 5.1-5.4 marked as
superseded by Option A (anonymous flow stays agent-driven),
others closed against where they were actually completed.

Build clean, 186/186 tests pass.

Closes TSKL-231.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Copilot AI review requested due to automatic review settings May 11, 2026 06:25
Copilot AI reviewed May 11, 2026
View reviewed changes

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

This PR consolidates Taskless’s agent surface area by replacing multiple per-task skills/commands with a single taskless router skill and a single /tskl router command, shifting detailed agent instructions into on-demand CLI “help recipes” (taskless help <topic>). It also updates the CLI to support recipe variant lookup (--anonymous), standardized JSON error envelopes, a rules → rule verb rename, and adds a dedicated taskless update subcommand.

Changes:

  • Replace 10 Taskless skills + 6 slash commands with one router skill (skills/taskless) and one router command (commands/tskl/tskl.md).
  • Add/expand taskless help recipe system (topic index, per-topic recipes, anonymous variants, schema interpolation).
  • Update CLI behavior: global --anonymous matrix, standardized { ok:false, code, message } JSON failures, telemetry taxonomy changes, and taskless update.

Reviewed changes

Copilot reviewed 100 out of 100 changed files in this pull request and generated 10 comments.

Show a summary per file
File Description
skills/taskless/SKILL.md Adds the consolidated router skill that delegates to taskless help <topic>.
skills/taskless-logout/SKILL.md Removes legacy per-task skill (logout).
skills/taskless-login/SKILL.md Removes legacy per-task skill (login).
skills/taskless-info/SKILL.md Removes legacy per-task skill (info).
skills/taskless-improve-rule/SKILL.md Removes legacy per-task skill (improve rule).
skills/taskless-improve-rule-anonymous/SKILL.md Removes legacy per-task skill (anonymous improve).
skills/taskless-delete-rule/SKILL.md Removes legacy per-task skill (delete rule).
skills/taskless-create-rule/SKILL.md Removes legacy per-task skill (create rule).
skills/taskless-create-rule-anonymous/SKILL.md Removes legacy per-task skill (anonymous create).
skills/taskless-check/SKILL.md Removes legacy per-task skill (check).
README.md Updates repo docs to describe the consolidated skill/command + recipe approach.
packages/cli/test/wizard-integration.test.ts Updates wizard integration tests for “no optional skills” and consolidated install output.
packages/cli/test/verify.test.ts Removes CLI-level rules verify --schema test now that --schema is removed.
packages/cli/test/rule-from.test.ts Updates tests for rules → rule verb rename.
packages/cli/test/init-no-interactive.test.ts Updates init tests for consolidated skill; adds taskless update coverage.
packages/cli/test/help-extensions.test.ts Adds end-to-end tests for help index/topics, interpolation, variants, and non-TTY routing.
packages/cli/test/error-envelope.test.ts Adds tests asserting standardized JSON error envelope codes/shape.
packages/cli/test/cli.test.ts Updates CLI install test to check new consolidated skill + router command.
packages/cli/test/anonymous-flag.test.ts Adds coverage for the per-command --anonymous behavior matrix.
packages/cli/src/wizard/steps/optional-skills.ts Removes optional-skills wizard step implementation.
packages/cli/src/wizard/index.ts Updates wizard flow to remove optional skills selection and install embedded skills directly.
packages/cli/src/util/schema-output.ts Removes deprecated --schema printing utility.
packages/cli/src/types/errors.ts Introduces stable error-code union + JSON envelope helpers.
packages/cli/src/install/catalog.ts Shrinks skill catalog to a single mandatory taskless skill.
packages/cli/src/index.ts Adds update subcommand, renames rules → rule, and routes non-TTY bare taskless to help index.
packages/cli/src/help/update.txt Adds a recipe for the new update topic.
packages/cli/src/help/rules.txt Removes old rules help topic file.
packages/cli/src/help/rules-verify.txt Removes old rules verify help topic file.
packages/cli/src/help/rules-meta.txt Removes old rules meta help topic file.
packages/cli/src/help/rules-improve.txt Removes old rules improve help topic file.
packages/cli/src/help/rules-delete.txt Removes old rules delete help topic file.
packages/cli/src/help/rules-create.txt Removes old rules create help topic file.
packages/cli/src/help/rule.txt Adds new rule umbrella recipe topic.
packages/cli/src/help/rule-verify.txt Adds new rule verify recipe topic.
packages/cli/src/help/rule-meta.txt Adds new rule meta recipe topic.
packages/cli/src/help/rule-improve.txt Adds new API-backed rule improve recipe topic.
packages/cli/src/help/rule-improve.anonymous.txt Adds local-only rule improve --anonymous recipe variant.
packages/cli/src/help/rule-delete.txt Adds new rule delete recipe topic.
packages/cli/src/help/rule-create.txt Adds new API-backed rule create recipe topic.
packages/cli/src/help/rule-create.anonymous.txt Adds local-only rule create --anonymous recipe variant.
packages/cli/src/help/init.txt Rewrites init help into the canonical recipe template.
packages/cli/src/help/info.txt Rewrites info help into the canonical recipe template, including --anonymous behavior.
packages/cli/src/help/ci.txt Adds CI wiring recipe as a help topic (replacing optional CI skill).
packages/cli/src/help/check.txt Rewrites check help into the canonical recipe template.
packages/cli/src/help/auth.txt Rewrites auth help into the canonical recipe template.
packages/cli/src/help/auth-logout.txt Removes old auth-logout help topic file.
packages/cli/src/help/auth-login.txt Removes old auth-login help topic file.
packages/cli/src/commands/init.ts Adds update subcommand and makes --anonymous accepted for init/update.
packages/cli/src/commands/info.ts Adds --anonymous behavior and completion telemetry; swaps to standardized envelope helper.
packages/cli/src/commands/help.ts Implements recipe lookup, anonymous variants, interpolation, and new telemetry events.
packages/cli/src/commands/check.ts Accepts --anonymous (no-op), adds standardized error envelope on scan failures, and completion telemetry.
packages/cli/src/commands/auth.ts Adds --anonymous behavior handling + completion telemetry.
packages/cli/README.md Updates CLI docs for singular rule, help recipes, and --anonymous behavior matrix.
openspec/specs/skill-taskless/spec.md Adds spec for the consolidated taskless skill/router behavior.
openspec/specs/skill-improve-rule/spec.md Removes superseded per-skill spec (improve rule).
openspec/specs/skill-delete-rule/spec.md Removes superseded per-skill spec (delete rule).
openspec/specs/skill-ci/spec.md Removes superseded per-skill spec (CI).
openspec/specs/skill-auth-logout/spec.md Removes superseded per-skill spec (auth logout).
openspec/specs/skill-auth-login/spec.md Removes superseded per-skill spec (auth login).
openspec/specs/cli/spec.md Extends CLI spec for --anonymous and standardized error envelopes.
openspec/specs/cli-flag-schema/spec.md Removes spec for the deprecated --schema flag.
openspec/specs/cli-check/spec.md Adds check requirements for --anonymous no-op + standardized envelope on failure.
openspec/specs/cli-auth/spec.md Updates auth spec for --anonymous handling and JSON envelope requirements.
openspec/specs/analytics/spec.md Updates telemetry taxonomy (help intent + action start/completion).
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/skills/spec.md Archives consolidated skill/command requirements snapshot (change record).
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/skill-taskless/spec.md Archives consolidated taskless skill spec (change record).
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/skill-improve-rule/spec.md Archives removal/migration notes for improve-rule skill spec.
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/skill-delete-rule/spec.md Archives removal/migration notes for delete-rule skill spec.
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/skill-create-rule/spec.md Archives removal/migration notes for create-rule skill spec.
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/skill-ci/spec.md Archives removal/migration notes for CI skill spec.
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/skill-auth-logout/spec.md Archives removal/migration notes for auth-logout skill spec.
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/skill-auth-login/spec.md Archives removal/migration notes for auth-login skill spec.
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/cli/spec.md Archives CLI requirements for consolidation (change record).
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/cli-init/spec.md Archives init/update behavior requirements (change record).
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/cli-help/spec.md Archives help recipe/variant/schema requirements (change record).
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/cli-flag-schema/spec.md Archives removal/migration notes for --schema flag.
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/cli-check/spec.md Archives check changes requirements (change record).
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/cli-auth/spec.md Archives auth changes requirements (change record).
openspec/changes/archive/2026-05-10-consolidate-taskless-skills/specs/analytics/spec.md Archives telemetry taxonomy changes (change record).
commands/tskl/tskl.md Adds single /tskl router command file.
commands/tskl/rule.md Removes legacy per-command router file.
commands/tskl/logout.md Removes legacy per-command router file.
commands/tskl/login.md Removes legacy per-command router file.
commands/tskl/info.md Removes legacy per-command router file.
commands/tskl/improve.md Removes legacy per-command router file.
commands/tskl/check.md Removes legacy per-command router file.
.changeset/consolidate-taskless-skills.md Adds changeset describing the consolidation, breaking changes, and new features.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Comment thread skills/taskless/SKILL.md
Comment thread commands/tskl/tskl.md
Comment thread packages/cli/src/help/rule-create.txt Outdated
Comment thread packages/cli/src/help/rule-create.anonymous.txt Outdated
Comment thread packages/cli/src/help/rule-create.anonymous.txt Outdated
Comment thread packages/cli/src/help/rule-delete.txt
Comment thread packages/cli/src/help/check.txt Outdated
Comment thread packages/cli/src/commands/info.ts
Comment thread packages/cli/src/help/auth.txt
Comment thread packages/cli/src/commands/auth.ts
thecodedrift and others added 2 commits May 11, 2026 08:52
@thecodedrift @claude
fix(cli): Unbreak CI build and address PR #18 review feedback
833dcd7 
Drop obsolete scripts/generate-commands.ts and its build:/bump: hooks.
The script's prefix-based "skills/taskless-*" model is incompatible
with the v0.7 consolidation: there is now one skills/taskless/SKILL.md
plus a single hand-authored commands/tskl/tskl.md whose body
intentionally diverges from the skill body. Update the infrastructure
spec and README to reflect that command files are now hand-maintained.

Address Copilot review feedback on PR #18:

- rule-create.txt: report timestamped test path
  .taskless/rule-tests/<id>-YYYYMMDD-test.yml in step 9.
- rule-create.anonymous.txt: stop claiming {{INPUT_SCHEMA}} embeds the
  ast-grep schema. Point step 1 to the upstream ast-grep rule
  reference; rewrite the "ast-grep schema" section to be honest about
  what is and isn't embedded; reword the Errors header so verify's
  layered output is not confused with the {ok:false,...} envelope.
- rule-delete.txt: replace the --json envelope table with the actual
  plain-text behavior; rule delete does not accept --json.
- auth.txt: replace the --json envelope table with documented
  plain-text behavior for each failure mode; auth commands do not
  accept --json.
- check.txt: clarify that success:false in JSON output means rule
  matches were found (an expected outcome) and that the
  {ok:false,code,message} envelope only applies to scan failures
  such as SCAN_FAILED.
- info.ts: write the INTERNAL_ERROR envelope to stdout via console.log
  instead of stderr, matching every other JSON-mode output path.

Lint, typecheck, build, and the 186-test suite all pass locally.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
feat(cli): Add --json envelope support to auth and rule delete
24c6063 
The cli capability spec mandates a standardized { ok:false, code,
message } envelope on stdout when --json is set and the command
exits with an error. cli-auth/spec.md scenario "Auth login network
failure in JSON mode" makes this explicit for the auth subcommands.
The implementations were missing the flag entirely, leaving the
recipes and the spec divergent.

Add --json support and the envelope path to:

- auth login: --anonymous rejection emits INVALID_INPUT; device-flow
  cancellations route by reason (denied -> AUTH_REQUIRED; expired or
  caught error -> NETWORK_ERROR). The URL/code chatter is suppressed
  in --json mode so stdout stays a single parseable envelope on
  failure.
- auth logout: accepts --json. Success is silent on stdout; the
  envelope path is reserved for future error cases. No real error
  paths exist today, so no behavior change on the happy path.
- auth (status, no subcommand): accepts --json for forward-compat;
  status output stays plain text since there are no error paths.
- rule delete: --json + missing rule -> RULE_NOT_FOUND envelope.
  Success is silent on stdout in --json mode.

Recipes (auth.txt, rule-delete.txt) are restored to document the
--json envelope contract now that the implementation matches.

Tests in error-envelope.test.ts cover rule delete (RULE_NOT_FOUND
and silent success), auth login --anonymous --json (INVALID_INPUT),
and auth logout --json (silent success). 190 tests pass.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift
thecodedrift requested a review from Copilot May 11, 2026 17:58
Copilot AI reviewed May 11, 2026
View reviewed changes

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Copilot reviewed 103 out of 103 changed files in this pull request and generated 5 comments.

Comment thread packages/cli/src/help/check.txt
Comment thread packages/cli/src/help/check.txt Outdated
Comment thread README.md Outdated
Comment thread packages/cli/src/commands/auth.ts
Comment thread packages/cli/src/help/update.txt Outdated
thecodedrift and others added 2 commits May 11, 2026 14:51
@thecodedrift @claude
chore: Migrate local .claude install to consolidated taskless skill
e077854 
Result of running the locally-built CLI to install the v0.7
consolidation into this repo's own .claude directory, used as a test
bed for verifying the install/migration flow end-to-end.

- Remove the 10 stale per-task taskless-* skill symlinks
  (taskless-check, taskless-ci, taskless-create-rule,
  taskless-create-rule-anonymous, taskless-delete-rule,
  taskless-improve-rule, taskless-improve-rule-anonymous,
  taskless-info, taskless-login, taskless-logout). Their source
  directories were deleted in the consolidation; the symlinks were
  dangling.
- Add .claude/skills/taskless symlink pointing at the consolidated
  skill source.
- Add .claude/.gitignore to keep local-state files (*.local.json,
  *.lock) out of the repo.
- Update .taskless/taskless.json to reflect the new single-skill
  manifest and bump the recorded cliVersion to 0.6.0.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift @claude
fix(cli): Address PR #18 review round 4 (check semantics, telemetry, …
8d18b68 
…recipes)

Address the latest Copilot review feedback on PR #18:

- check.txt (medium): correct the `success` field semantics. The CLI
  sets success based on whether any error-severity finding exists,
  not whether any match exists. Rewrite step 4 to explicitly cover
  all three cases: success:false (>=1 error finding, exit 1);
  success:true with non-empty results (warning/info/hint only,
  exit 0); empty results (clean). Restate that findings are not
  envelope-routed; the envelope only appears on scan failure.
- check.txt (low): rewrite the step-3 JSON sample to match the real
  `CheckResult` shape (`source`, `ruleId`, `severity`, `message`,
  `note`, `file`, `range: { start, end }`, `matchedText`, `fix`) so
  agents don't parse non-existent keys.
- README.md (low): add `update` to the `taskless help` topic
  inventory now that the recipe and subcommand exist.
- auth.ts (low): track the emitted `CliErrorCode` in the login
  command's `fail` closure and include `errorCode` on
  `cli_auth_login_completed` when `success: false`, per the
  analytics spec. Scoped to this file; rolling the pattern across
  other `cli_<action>_completed` events is a broader cleanup
  tracked separately.
- update.txt (low): drop the contradictory "silent on success" line
  and document the actual behavior (prints a summary of
  skills/commands added, removed, or kept in sync, matching
  `init --no-interactive`).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift
thecodedrift requested a review from Copilot May 12, 2026 00:54
Copilot AI reviewed May 12, 2026
View reviewed changes

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Copilot reviewed 116 out of 116 changed files in this pull request and generated 1 comment.

Comment thread packages/cli/src/help/rule-create.anonymous.txt Outdated
@thecodedrift @claude
docs(cli): Clarify rule verify --json output shapes in anonymous recipe
43b8871 
The Errors section of rule-create.anonymous.txt claimed
\`rule verify --json\` always emits the layered result and never the
\`{ ok:false, code, message }\` envelope. That's wrong: when \`<id>\`
is missing, the command emits the standardized INVALID_INPUT
envelope on stdout and exits non-zero; only when \`<id>\` is supplied
does it emit the layered \`{ success, schema, requirements, tests }\`
shape. Split the Errors table into envelope vs layered rows so the
two paths are unambiguous.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@thecodedrift
thecodedrift marked this pull request as ready for review May 12, 2026 03:08
@thecodedrift
thecodedrift merged commit af454cd into main May 12, 2026
2 checks passed
thecodedrift added a commit that referenced this pull request May 12, 2026
@thecodedrift @claude
fix(cli): Unbreak CI build and address PR #18 review feedback
cd94356 
Drop obsolete scripts/generate-commands.ts and its build:/bump: hooks.
The script's prefix-based "skills/taskless-*" model is incompatible
with the v0.7 consolidation: there is now one skills/taskless/SKILL.md
plus a single hand-authored commands/tskl/tskl.md whose body
intentionally diverges from the skill body. Update the infrastructure
spec and README to reflect that command files are now hand-maintained.

Address Copilot review feedback on PR #18:

- rule-create.txt: report timestamped test path
  .taskless/rule-tests/<id>-YYYYMMDD-test.yml in step 9.
- rule-create.anonymous.txt: stop claiming {{INPUT_SCHEMA}} embeds the
  ast-grep schema. Point step 1 to the upstream ast-grep rule
  reference; rewrite the "ast-grep schema" section to be honest about
  what is and isn't embedded; reword the Errors header so verify's
  layered output is not confused with the {ok:false,...} envelope.
- rule-delete.txt: replace the --json envelope table with the actual
  plain-text behavior; rule delete does not accept --json.
- auth.txt: replace the --json envelope table with documented
  plain-text behavior for each failure mode; auth commands do not
  accept --json.
- check.txt: clarify that success:false in JSON output means rule
  matches were found (an expected outcome) and that the
  {ok:false,code,message} envelope only applies to scan failures
  such as SCAN_FAILED.
- info.ts: write the INTERNAL_ERROR envelope to stdout via console.log
  instead of stderr, matching every other JSON-mode output path.

Lint, typecheck, build, and the 186-test suite all pass locally.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
thecodedrift added a commit that referenced this pull request May 12, 2026
@thecodedrift @claude
fix(cli): Address PR #18 review round 4 (check semantics, telemetry, …
3de550c 
…recipes)

Address the latest Copilot review feedback on PR #18:

- check.txt (medium): correct the `success` field semantics. The CLI
  sets success based on whether any error-severity finding exists,
  not whether any match exists. Rewrite step 4 to explicitly cover
  all three cases: success:false (>=1 error finding, exit 1);
  success:true with non-empty results (warning/info/hint only,
  exit 0); empty results (clean). Restate that findings are not
  envelope-routed; the envelope only appears on scan failure.
- check.txt (low): rewrite the step-3 JSON sample to match the real
  `CheckResult` shape (`source`, `ruleId`, `severity`, `message`,
  `note`, `file`, `range: { start, end }`, `matchedText`, `fix`) so
  agents don't parse non-existent keys.
- README.md (low): add `update` to the `taskless help` topic
  inventory now that the recipe and subcommand exist.
- auth.ts (low): track the emitted `CliErrorCode` in the login
  command's `fail` closure and include `errorCode` on
  `cli_auth_login_completed` when `success: false`, per the
  analytics spec. Scoped to this file; rolling the pattern across
  other `cli_<action>_completed` events is a broader cleanup
  tracked separately.
- update.txt (low): drop the contradictory "silent on success" line
  and document the actual behavior (prints a summary of
  skills/commands added, removed, or kept in sync, matching
  `init --no-interactive`).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

Copilot code review Copilot
Copilot review effort
Applies to this pull request for everyone.Learn more about Copilot code review.
Copilot left review comments

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@thecodedrift