agentleFS
alekspetrovalekspetrov/navigator/skills/nav-brief/SKILL.md

nav-brief

Skill356 starsChanged 13 months agoMITView on GitHub

Render a one-screen intent brief (Goal/Scope/Approach/Limits/Verify/Won't-do/Contradiction) before implementing ambiguous task-shaped prompts, triggered by the prompt_brief op (UserPromptSubmit via hooks/nav_dispatch.py). Confirms scope with max 2 open questions before touching files; detects brief drift mid-task.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use the template below. Pre-fill every field you can from the injected ## Relevant Memories section and current session context.

    line 40
  2. Distinctive:

    Max 2 open questions: Ask only what you cannot infer or safely assume.

    line 65
  3. Very specific:

    After confirmation, capture corrections: changed defaults are candidate knowledge-graph memories ("Remember we decided..." flow via …

    line 69
  4. Very specific:

    When the row is not none, query prior resolutions BEFORE filling Approach:

    line 80
  5. Very specific:

    nav-triz: before filling Approach: it produces three candidates from different separation modes and recommends one.

    line 93

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Navigator Intent Brief Skill

Front-load the scope negotiation. When a prompt is task-shaped but ambiguous, render a one-screen INTENT BRIEF and get confirmation before writing code — turning 2-4 correction exchanges into 1 confirmation exchange.

Why This Exists

Ambiguity is not complexity. A small task can be highly ambiguous ("fix the bug" — which bug?) while a large one can be fully specified. The expensive failure mode is implementing on guessed scope: rework burns tokens on undo/redo, pollutes context with dead ends, and forces mid-feature compacts.

When This Fires

Hook contract: render a brief ONLY when this turn's injected context contains a NAV-BRIEF reminder block (emitted by the prompt_brief op — hooks/ops/prompt_brief.py, dispatched by nav_dispatch.py — on UserPromptSubmit when ambiguity score >= threshold). Never self-trigger from your own judgment alone — the hook is the single trigger source, so behavior stays predictable and tunable via config.

Config (.agent/.nav-config.json):

"brief_hook": {
  "enabled": true,
  "ambiguity_threshold": 0.5,
  "memory_budget_chars": 1200
}

Rendering the Brief

Use the template below. Pre-fill every field you can from the injected ## Relevant Memories section and current session context; mark assumptions explicitly.

The brief is a bullet list, one line per field. Claude Code Desktop draws replies in a proportional face, so a box of │ and ─ outside a code fence falls apart there (TASK-108). Never draw box glyphs in the reply itself.

**BRIEF: <short title>**
- **Goal** <outcome, one line>
- **Scope** <files/dirs/endpoints — concrete>; NOT <adjacent things left untouched>
- **Approach** <how — cite memory if pre-filled>
- **Limits** <numbers, constraints — mark ASSUMED>
- **Verify** <how completion will be proven>
- **Won't do** <explicit non-goals>
- **Contradict** <improving X worsens Y — or "none">

Confirm / edit? (open questions: <0-2>)

A fenced text block holding the older box form is the one alternative: inside a fence both surfaces draw a monospace grid, so the box keeps its edges.

Rules:

  • Max 2 open questions. Ask only what you cannot infer or safely assume; everything else is a pre-filled default the user can veto.
  • Stop and wait for user confirmation before any Edit/Write/Bash that modifies the project.
  • After confirmation, capture corrections: changed defaults are candidate knowledge-graph memories ("Remember we decided..." flow via nav-graph), so future briefs pre-fill better.

When a Contradiction Is Declared (TRIZ, TASK-72)

Contradict is the TRIZ row: the task is inventive only when improving one thing worsens another ("faster session start vs enough context"). Most tasks are routine — none is the expected common case. Never invent a tension to fill the row.

When the row is not none, query prior resolutions BEFORE filling Approach:

PLUGIN_DIR="${CLAUDE_PLUGIN_ROOT:-$(cat "${NAVIGATOR_CONFIG_HOME:-${XDG_CONFIG_HOME:-$HOME/.config}/navigator}/plugin-root" 2>/dev/null)}"
[ -d "$PLUGIN_DIR/skills" ] || PLUGIN_DIR="$HOME/.claude/plugins/marketplaces/navigator-marketplace"
python3 "$PLUGIN_DIR/skills/nav-graph/functions/graph_manager.py" \
  --action contradictions --filter "<2-3 key terms>"

Cite matching mem-NNN ids in Approach ("resolved by separation in time before: mem-063"). If nothing matches, say so in one line. On a substantial task (Task Mode complexity ≥ 0.5 or more than one subsystem), hand off to nav-triz before filling Approach: it produces three candidates from different separation modes and recommends one. After the task, if the resolution was non-obvious, capture it as a decision with --contradiction / --separation / --principle (see nav-graph Step 3B) so the next brief can find it. Drift detection does not cover this row.

Passthrough Rules (do NOT render a brief)

  • No NAV-BRIEF reminder in this turn's context.
  • The prompt is plainly an answer to a pending brief's open questions (model-side judgment — the hook is stateless in v1 and may re-fire on answers that contain task verbs; known limitation).
  • User signals urgency/override: "just do it", "quick fix", "skip the brief". Respect it silently.

Brief Drift Detection

Mid-implementation, if the work is about to exceed the confirmed brief's Scope or Limits (touching files outside scope, changing an untouched-by- agreement area), STOP and surface:

⚠ BRIEF DRIFT: <what exceeds the brief, one line>
Options: extend brief / skip this change

One question, then continue per the answer.

Interaction with workflow_enforcer / Task Mode

Both UserPromptSubmit hooks may fire on the same prompt. Show both blocks — they are complementary: WORKFLOW CHECK decides how to execute (mode), the brief decides what is in scope. In Task Mode, the confirmed brief replaces ad-hoc scope derivation in the PLAN phase.

Examples

Brief rendered — "add rate limiting to the API" (score 0.7): brief with Scope pre-filled to public routes (graph memory), Limits marked ASSUMED, 1 open question about limit numbers. User answers; implementation starts.

Passthrough — "fix the typo in README.md" (score 0.1): file reference gives full scope; no brief, direct execution.

Drift — brief confirmed scope /api/public/*; implementation later needs a shared middleware file used by internal routes → BRIEF DRIFT raised, user extends the brief.

Future Work (explicitly deferred from v1)

  • Statefulness: persist pending-brief state across turns (removes the answer-re-fire limitation).
  • strict_block: hard-gate implementation until a brief is confirmed.
  • Measurement: briefs-shown/passthrough counters + corrections-per-task metrics (pairs with nav-stats; deferred until there is a consumer).
  • Prompt rewriting.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.